Two experiments are reported that tested core assumptions of the mental models theory of syllogistic inference (e.g., Johnson-Laird & Byrne, 1991) by examining inspection times for syllogistic components. Results supported mental models predictions of: (1) increased cognitive load across syllogistic figures, with differences in processing demand arising for BA-CB versus AB-BC problems for both conclusionevaluation and conclusion-production tasks; and (2) conclusion-order preferences across figures – again in both the evaluation and the production paradigms. These findings challenge views of figural biases as being confined to conclusion-production tasks (Geurts, 2003; Rips, 1994) and theories that reject the assumption of figure-induced cognitive load (Chater & Oaksford, 1999). Since figural effects are typically viewed as being indicative of premise-driven processing, these results are also inconsistent with proposals that premise-driven processing prevails in conclusionproduction, and conclusion-driven processing dominates in conclusion-evaluation (Morley, Evans, & Handley, 2004). The results also clarify the role of processing demands associated with conclusion validity: Valid conclusions were scrutinised less than invalid ones in the evaluation paradigm (as predicted by Hardman & Payne, 1995), supporting the notion that invalid syllogisms have at least two mental models. Although our specific results are not entirely consistent with recent models-based proposals, our basic findings remain broadly compatible with a models approach rather than alternative theoretical positions.