This study explored the abilities of 5th, 8th, and 10th graders, and College students to reason logically about what they read. Both students' metacognitive behavior (looking back at previously read text) and their performance on logical deduction questions were recorded and analyzed in a reading task. Conditional logic premises and deductive questions were embedded in three narratives containing premise information that was factual (True Story), contrary to fact (False Story), or unverifiable via common world knowledge (Neutral Story). The texts and questions were presented one sentence at a time on a computer screen; participants controlled the presentation of sentences. For answering the questions, three response tasks were devised. One task (labeled Generate) required readers to generate their own logical conclusions in response to deduction questions. Two tasks (labeled Valid and Invalid) required readers to evaluate logically valid or logically invalid conclusions drawn by story characters in the texts. Students in early and late adolescence looked back more when asked to evaluate logical conclusions than when asked to generate conclusions on their own; College students' lookback frequencies were not significantly affected by response task, but were greater overall than those of younger students. With conditional forms requiring an uncertainty response (Affirmed Consequent and Denied Antecedent), readers looked back more when evaluating logically invalid conclusions than when evaluating logically valid ones. Readers of all ages were more likely to agree with story characters' (valid) uncertain conclusions with the AC and DA forms than they were to disagree with story characters' (invalid) certain conclusions to these forms. Both lookback frequency and performance on logic questions were lowest when readers were required to reason from contrary to fact premises.