Data from 31 countries participating in the Programme for International Student Assessment (PISA) is used to estimate education production functions for reading literacy. The analysis suggests that the probability of finding statistically significant and correctly signed class size effects increases the higher the level of aggregation used to measure class size.