Valve-point loading affects the input-output characteristics of generating units, bringing the fuel costs nonlinear and nonsmooth. This has been considered in the solution of load dispatch problems, but not in the planning phase of unit commitment. This paper presents a mathematical optimization model for the thermal unit commitment problem considering valve-point loading. The formulation is based on a careful linearization of the fuel cost function, which is modeled with great detail on power regions being used in the current solution, and roughly on other regions. A set of benchmark instances for this problem is used for analyzing the method, with recourse to a general-purpose mixed-integer optimization solver.