As part of an ongoing project to examine student learning in upper-division courses in thermal and statistical physics, we have examined student reasoning about entropy and the second law of thermodynamics. We have examined reasoning in terms of heat transfer, entropy maximization, and statistical treatments of multiplicity and probability. In this paper, we describe student responses in interviews focused on the approach of macroscopic systems to thermal equilibrium. Our data suggest that students do not use a single simple model of entropy, but rather use a variety of conceptual resources. Individual students frequently shifted between resources, in some cases leading to contradictory predictions. Among the resources that students employed were some that have been previously described in the literature, including inappropriate use of conservation. However, our results suggest that student use of resources connected to disorder are neither simple nor monolithic. For example, many students used a previously unreported association between the equilibrium state of a system and an increase in order, rather than disorder. [This paper is part of the Focused Collection on Upper Division Physics Courses.]