A model demonstrates how low correlations for academic achievement in separate units of instruction can be compatible with high correlations for cumulated achievement. The unit achievement is assumed to consist of a task-specific part and a part correlated with a predictor. When units of learning are cumulated, attrition of earlier learning takes place according to a certain retention ratio. This paper illustrates how correlations with achievement in other tasks and with a predictor are affected by the cumulation process, using various patterns of parameter values. (Author/MV)