We investigate Ornstein-Uhlenbeck and diffusion processes with variable drift rates as models of evidence accumulation in a visual discrimination task. We derive power-law and exponential drift-rate models and characterize how parameters of these models affect the psychometric function describing performance accuracy as a function of stimulus strength and viewing time. We fit the models to psychophysical data from monkeys learning the task to identify parameters that best capture performance as it improves with training. The most informative parameter was the overall drift rate describing the signal-to-noise ratio of the sensory evidence used to form the decision, which increased steadily with training. In contrast, secondary parameters describing the time course of the drift during motion viewing did not exhibit steady trends. The results indicate that relatively simple versions of the diffusion model can fit behavior over the course of training, thereby giving a quantitative account of learning effects on the underlying decision process.