Due to the dramatic difference in spatial resolution between the central fovea and the surrounding retinal regions, accurate fixation on important objects is critical for humans. It is known that the preferred retinal location (PRL) for fixation of healthy human observers rarely coincides with the retinal location with the highest cone density. It is not currently known, however, whether the PRL is consistent within an observer or is subject to fluctuations and, moreover, whether observers’ subjective fixation location coincides with the PRL. We studied whether the PRL changes between days. We used an adaptive optics scanning laser ophthalmoscope to project a Maltese cross fixation target on an observer's retina and continuously imaged the exact retinal location of the target. We found that observers consistently use the same PRL across days, regardless of how much the PRL is displaced from the cone density peak location. We then showed observers small stimuli near the visual field location on which they fixated, and the observers judged whether or not the stimuli appeared in fixation. Observers’ precision in this task approached that of fixation itself. Observers based their judgment on both the visual scene coordinates and the retinal location of the stimuli. We conclude that the PRL in a normally functioning visual system is fixed, and observers use it as a reference point in judging stimulus locations.

*SD*, 1.79, excluding the exceptional 206-day separation).

*SD*, 234.9) per day per observer.

*SD*, 4.39), and there were no cases of skewness over 1 (mean, 0.54;

*SD*, 0.28). Thus, because the analysis of the data of Experiment 2 and the comparison between results from the two experiments rely very much on Gaussian distributions, we also fitted the data from Experiment 1 with 2D Gaussian distributions. However, to make sure that the PRLs we are reporting were not specific to a particular fitting method, we also calculated (distribution assumption–free) geometric medians. The location parameters derived in the two ways are in very good agreement, differing on average less than a foveal cone diameter (mean, 0.142 arcmin;

*SD*, 0.093; range, 0.03–0.39) (Supplementary Table S1). Differences in the PRLs between two days were analyzed for each observer with the Akaike information criterion (AIC). For observers with three measurements, the two with the largest location difference were used. The AIC values were obtained by fitting 2D Gaussian distributions to the data by means of maximum likelihood estimation with the MATLAB fitgmdist function (MathWorks, Natick, MA). In the simpler model (m1), fixation data from 2 days were fitted with shared parameters only. In the more complex model (m2), the mean (

*x*,

*y*) coordinates of the two distributions were allowed to differ (shared covariance). The evidence ratio was then calculated from the difference between the AIC values for the more simple model (m1) and the more complex model (m2) as

*P*(m1 is best)/

*P*(m2 is best), where

*P*(m1 is best) = exp(–ΔAIC/2)/[1 + exp(–ΔAIC/2)] and

*P*(m2 is best) = 1 –

*P*(m1 is best). We additionally analyzed the difference between the same 2 days with a completely distribution assumption–free, two-sample, 2D Kolmogorov–Smirnov (KS) test (Fasano & Franceschini, 1987) with the MATLAB kstest2d function (Lau, 2016). To make sample observations within each day independent, as assumed by the KS test, stimulus locations were averaged across each 2- to 6-second fixation epoch. The range of

*D*-statistics of the tests was 0.27 to 0.33 (mean, 0.30;

*n*= 25–30 per session per observer) and the range of

*p*values was 0.114 to 0.295 (mean, 0.201).

*x*and

*y*) center location parameters were forced to be the same for the three functions. The standard deviations of the different distributions were allowed to vary independently. The SFL was then given by the common center location parameters of the fitted distributions. The two red ellipses in Figures 3B and 3C correspond to the regions where the DoG function reached 60.6% of maximum (corresponding to 1

*SD*distance from the peak in a single Gaussian function) on the way to the central dip (smaller red ellipse) and to the outer plateau (larger red ellipse).

*t*-test. Because the distance between the PRL and the cone peak affects how much closer the SFL can be to one of them (if the PRL and the cone peak are at the same location, for example, then the SFL is unavoidably equally far from both), we normalized both the PRL–SFL distance and the cone peak–SFL distance with the PRL–cone peak distance before the

*t*-test.

*r*= 0.47;

*p*< 0.001). Crucially, though, they also have a significant amount of independent variance. A generalized linear mixed model was used to analyze the contributions of the raster location and the retinal location of the stimulus on the probability of different response alternatives. Because the response variable was an ordinal-scale variable with three possible values, a multinomial logistic link function was used. The analysis was carried out with SPSS Statistics 25 (IBM, Armonk, NY). The repeated-measures nature of the data was incorporated by adding observer as a random factor into the model. In addition, we tested for differences in the proportions of “yes” responses in two circular regions of interest (area coincident with the PRL and an area displaced from the raster center toward a direction opposite the PRL) with a chi-square test for independence, with the following sample sizes: 10003R,

*n*= 63 (PRL) and

*n*= 43 (opposite); 20094R,

*n*= 205 (PRL) and

*n*= 225 (opposite); and 20109R,

*n*= 160 (PRL) and

*n*= 170 (opposite). Sample sizes differed considerably between observers, because (to limit the amount of overlap between regions of interest) the diameter of the analyzed region for each condition was equal to the distance between the raster center and the PRL, in retinal coordinates, which differed among subjects.

*SD*, 0.127).

*SD*, 4.32), very close to the 5.62-arcmin earlier observed by Li et al. (2010). We also calculated the sampling frequency limits at the cone density peak and the PRL, assuming hexagonal packing (Wang et al., 2019). The average sampling frequency limits were 71.9 cycles per degree (

*SD*, 2.48) for the cone density peak and 69.3 cycles per degree (

*SD*, 2.74) for the PRL.

*SD*over observers, 13,834) for the single Gaussian model and 120,262 (

*SD*, 13,834) for the two-Gaussian model. The evidence ratio derived from the AIC analysis suggests that the single Gaussian model was on average 22.8 (

*SD*, 4.1) times more likely to be the better model than the model with different average locations for the 2 days. Because our data were not strictly Gaussian, we also conducted two-sample KS tests on the same datasets (see Quantification and Statistical Analysis section), which, in line with the AIC analysis, did not indicate a statistically significant difference in the PRL positions for any observers (

*p*> 0.1). Although not decisive evidence for either case, both analyses rather support a view of the same PRL across days.

*SD*from the mean. Figure 3D shows normalized densities (averaged over five observers) of different responses as a function of distance from the SFL. One can see that the bimodal function was a correct choice for modeling the “no” responses.

*SD*) of the SFL could have easily been affected by perceptual mislocalization due to microsaccades occurring near stimulus onset (Hafed, 2013). Considering this possibility, we also fitted the above-described functions to data where all trials with a microsaccade occurring within 250 ms from stimulus onset were filtered out. However, that had virtually no effect on either the location or the distribution of the SFL. Because of that, and the fact that microsaccades were not filtered from Experiment 1 data, we used only unfiltered data in all further analyses of Experiment 2.

*SD*, 1.93) than to the cone density peak (black cross; mean distance, 7.63 arcmin;

*SD*, 5.25) (Figure 4C). This distance difference was statistically significant,

*t*(4) = 5.35,

*p*= 0.006. Further, both the PRL and SFL were displaced from the cone density peak to a very similar polar angle direction (e.g., both upward for observer 20109R). The correlation between the directions to which the PRL and SFL were displaced from the cone density peak was quite high,

*r*(3) = 0.91,

*p*= 0.032.

*SD*ellipses was on average 5.54 arcmin (

*SD*, 0.730) for the “yes” responses and 3.91 arcmin (

*SD*, 1.21) for the fixation data. The PRL ellipse was smaller for all observers (see Figures 4A and 4B for examples). The difference was statistically significant,

*t*(4) = 5.46,

*p*= 0.006. Please note that, although the green ellipses in Figures 4A and 4B indicate the

*SD*of “yes” responses only, the location of the SFL (green markers) was determined based on all responses.

*F*(2, 5318) = 62.17,

*p*< 0.001, and the distance from the PRL,

*F*(2, 5318) = 34.17,

*p*< 0.001, were statistically significant. The interaction effect was not significant,

*F*(2, 5318 = 0.21,

*p*= 0.811. Figures 3D and 3E illustrate the drop in the probability of the stimulus being perceived as “in fixation” with increasing distance of the stimulus from raster center (black curve), with increasing distance of the stimulus from the PRL (magenta curve), and with the distance of the stimulus increasing simultaneously from the PRL and raster center (black-magenta curve).

*x*-axis) for whom the PRL and the average retinal location of the raster center were sufficiently apart that the above-described comparison is meaningful. In line with model predictions, stimuli displaced toward the PRL led to more “yes” responses than stimuli displaced to the opposite direction. The difference was statistically significant for each of the three observers: 10003R, χ

^{2}(1,

*N*= 106) = 7.57 and

*p*= 0.006; 20094R, χ

^{2}(1,

*N*= 430) = 5.88 and

*p*= 0.015; and 20109R, χ

^{2}(1,

*N*= 330) = 17.25 and

*p*< 0.001.

*Nature,*400(6747), 864–866, https://doi.org/10.1038/23693.

*Journal of Biomedical Optics,*13(2), 024008, https://doi.org/10.1117/1.2907211.

*BioRxiv,*https://doi.org/10.1101/2021.04.14.439841.

*Trends in Ecology & Evolution,*33(5), 358–372, https://doi.org/10.1016/j.tree.2018.03.001.

*Monthly Notices of the Royal Astronomical Society,*225(1), 155–170, https://doi.org/10.1093/mnras/225.1.155.

*Vision Research,*89, 65–71, https://doi.org/10.1016/j.visres.2013.07.011.

*Neuron,*77(4), 775–786, https://doi.org/10.1016/j.neuron.2012.12.014.

*Journal of Eye Movement Research,*12(6), 10, https://doi.org/10.16910/jemr.12.6.13.

*American Journal of Ophthalmology,*154(5), 767–778.e2, https://doi.org/10.1016/j.ajo.2012.05.007.

*Vision Research,*35(12), 1741–1754, https://doi.org/10.1016/0042-6989(94)00255-K.

*Current Biology,*23(17), 1663–1669, https://doi.org/10.1016/j.cub.2013.06.056.

*Investigative Ophthalmology & Visual Science,*51(12), 6858–6867, https://doi.org/10.1167/iovs.10-5499.

*Journal of the Optical Society of America. A, Optics, image science, and vision,*14(11), 2873–2883, https://doi.org/10.1364/JOSAA.14.002873.

*Vision Research,*41(5), 599–610, https://doi.org/10.1016/S0042-6989(00)00259-5.

*Journal of the Optical Society of America. A, Optics, image science, and vision,*14(4), 731–740, https://doi.org/10.1364/JOSAA.14.000731.

*Trends in Neurosciences,*32(9), 463–475, https://doi.org/10.1016/j.tins.2009.05.006.

*Visual psychophysics*(pp. 331–380). Heidelberg: Springer, https://doi.org/10.1007/978-3-642-88658-4_13.

*PLoS One,*13(11), e0207102, https://doi.org/10.1371/journal.pone.0207102.

*Biomedical Optics Express,*11(3), 1617–1632, https://doi.org/10.1364/BOE.384229.

*Retina,*20(2), 170–175, https://doi.org/10.1097/00006982-200002000-00010.

*Annual Review of Vision Science,*1(1), 569–594, https://doi.org/10.1146/annurev-vision-082114-035835.

*Biomedical Optics Express,*3(2), 240–258, https://doi.org/10.1364/BOE.3.000240.

*Vision Research,*118, 17–24, https://doi.org/10.1016/j.visres.2014.11.007.

*Current Biology,*23(17), 1691–1695, https://doi.org/10.1016/j.cub.2013.07.007.

*Nature Neuroscience,*20(10), 1413–1417, https://doi.org/10.1038/nn.4622.

*Journal of Refractive Surgery,*21(5), S575–S580, https://doi.org/10.3928/1081-597X-20050901-30.

*Progress in Retinal and Eye Research,*35, 63–81, https://doi.org/10.1016/j.preteyeres.2013.01.005.

*Journal of Vision,*5(7):3, 632–639, https://doi.org/10.1167/5.7.3.

*The effects of retinal image motion on the limits of spatial vision*(doctoral dissertation). Retrieved from https:///escholarship.org/uc/item/80v0d4z5.

*Optica Acta: International Journal of Optics,*16(2), 183–190, https://doi.org/10.1080/713818162.

*Current Biology,*https://doi.org/10.1016/j.cub.2021.07.005.

*Nature Neuroscience,*13(2), 156–157, https://doi.org/10.1038/nn.2465.

*Investigative Ophthalmology & Visual Science,*23(5), 666–670.

*Journal of the Optical Society of America,*55(9), 1158–1164, https://doi.org/10.1364/JOSA.55.001158.

*Proceedings of the 2010 Symposium on Eye-Tracking Research & Applications*(pp. 195–198). New York, NY: Association for Computing Machinery, https://doi.org/10.1145/1743666.1743714.

*European Journal of Neuroscience,*6(10), 1583–1588, https://doi.org/10.1111/j.1460-9568.1994.tb00548.x.

*Investigative Ophthalmology & Visual Science,*50(1), 84–89, https://doi.org/10.1167/iovs.08-2342.

*Optometry and Vision Science,*75(6), 399–406, https://doi.org/10.1097/00006324-199806000-00024.

*American Journal of Ophthalmology,*154(5), 779–789.e2, https://doi.org/10.1016/j.ajo.2012.05.004.

*Vision Research,*39(24), 4085–4097, https://doi.org/10.1016/S0042-6989(99)00108-X.

*Vision Research,*44(23), 2691–2705, https://doi.org/10.1016/j.visres.2004.05.027.

*Optics Express,*14(2), 487–497, https://doi.org/10.1364/OPEX.14.000487.

*Vision Research,*96, 75–86, https://doi.org/10.1016/j.visres.2014.01.005.

*eLife,*8, e47148, https://doi.org/10.7554/eLife.47148.

*Vision Research,*132, 53–61, https://doi.org/10.1016/j.visres.2016.05.001.

*Journal of Vision,*16(5):6, 1–10, https://doi.org/10.1167/16.5.6.