We develop a dynamic model of learning and wage determination: education may convey initial information about ability, but subsequent performance observations also are informative. Although the role of schooling in the labor market's inference process declines as performance observations accumulate, the estimated effect of schooling on the level of wages is independent of labor-market experience. In addition: time-invariant variables correlated with ability but unobserved by employers are increasingly correlated with wages as experience increases; wage residuals are a martingale; and wage cuts -are not rare, even for workers who do not change jobs. We present evidence from the National Longitudinal Survey of Youth that is generally consistent with all four of the model's predictions. We conclude that a blend of the learning model with an on-the-job-training model is more plausible than either model alone.
