Reduced chemical kinetics models are often used to lessen the computational cost of reacting flow simulations. However, because a reduced model is usually validated only for a nominal set of reaction conditions, unknown errors are introduced if the reduced model is used at new reaction conditions. In a previous paper, we introduced a method that, given a reduced model as input, identifies a rigorous range over which the model remains valid. However, this procedure is backwards: in most cases one starts with a known range of reaction conditions and one desires a reduced model that can be used over this range. Here we present the first automatic procedure that constructs a reduced chemistry model that is guaranteed to be valid everywhere in any user-specified range. The rigorousness of the model reduction method enables rigorous statements about the difference between the solution obtained using the reduced model and the solution that would have been obtained using the original full-chemistry model. By appropriate choice of error tolerances, the reduced-model solution can be made arbitrarily close to the full-model solution. This is demonstrated with adaptive chemistry simulations of one- and two-dimensional steady state laminar methane/air flames. As guaranteed by the error control procedure, the solutions of the reduced models are just as accurate as those obtained using the full-chemistry model, but they require significantly less CPU time.