Thirty-three participants viewed 1,000 pictures for 6 s each. Recognition was tested after 10 different intervals of time by mixing 100 of the original 1,000 with 100 new pictures. Participants judged each test picture “Old” or “New” on a 6-point scale. The unequal-variance recognition model is reinterpreted to estimate the probability of retrieval of an original (1,000) picture after each lapse of time. A second model then relates those different estimates of accessibility to the lapse of time, taking into account the interference on each test from pictures presented in preceding tests. Studies of category judgement explain (a) why the model distributions are normal, (b) why the operating characteristics are asymmetric, (c) why they are curvilinear, and (d) why the asymmetry decreases with lapse of time, this to justify a particular estimate of accessibility (probability of retrieval). Nine candidate functions are shown to the accessibilities. The underlying relation is a power law, but the exponent is poorly determined by the data (−1.5, −0.5), as also is the offset from the temporal origin. Comparisons with previous work identify two different relationships with respect to lapse of time: The retrieval of a unique image shows an approximately reciprocal loss, whereas a decrease in the amount of material reproduced by recall, recognition, or other method is approximately logarithmic. The present experiment exhibits both relationships, depending on whether specific account is taken of the effects of interference or, alternatively, interference is entirely ignored.