We present a formal analysis of symbolic learning that predicts significant differences in symbolic learning depending on the sequencing of semantic features and labels. A computational simulation confirms the Feature-Label-Ordering (FLO) effect in learning that our analysis predicts. Discrimination learning is facilitated when semantic features predict labels, but not when labels predict semantic features. A behavioral study confirms the predictions of the simulation. Our results and analysis suggest that the semantic categories people use to understand and communicate about the world might only be learnable when labels are predicted from objects.