The ability to correctly interpret data is an important skill in modern knowledge societies. The present study investigates adults' ability to interpret covariation data presented in bar graphs. Drawing on previous findings that show that the problem context influences the interpretation of contingency tables (grounded and concrete problems are easier than abstract ones) and based on findings from the literature on motivated reasoning (confirming problems are easier than disconfirming ones), we present N = 111 undergraduates with bar graphs in either grounded (confirming or disconfirming) or abstract contexts. Our results show that only grounded problems in confirming contexts are easier than abstract ones; grounded problems in disconfirming contexts are more challenging than abstract ones. Overall, the interpretation of bar graphs is difficult: Even in our sample of educated college students, correct performance did not exceed 50%. Our results support earlier findings regarding the context dependency of data-interpretation skills, and they suggest that relatively minor task variations have an impact on reasoners' interpretations of bar graphs.