Thinking about interactions between variables is a necessary condition for accurate scientific thinking. The purpose of this study was to investigate difficulties in thinking about interactions between variables and to suggest remedial educational means. A conceptual analysis distinguishes between valid interaction inferences, invalid interaction inferences, and limited inferences which can be seen as a partial or a primitive interaction inference. Empirical findings show that expert thinkers demonstrate thinking about interactions at both an operational and a metastrategic level. Some lay‐adults however, although able to draw many limited inferences, encounter substantial difficulties in drawing valid interaction inferences while engaged in an investigation of a scientific sort. Four types of difficulties were identified in this study: lack of a “double set of controlled comparisons” strategy that is necessary for valid inferences about interacting variables; lack of the conceptual framework for interacting factors; diversion of attention to other features; and difficulty in maintaining the necessary control of other variables. The implications of the findings to science curriculum are discussed.