The demand for lens systems of unconventional forms, coupled with the increased computing power available to most lens designers, has motivated the development of global optimization algorithms for lens design. The successful implementation of such an algorithm requires both a well designed global optimization scheme and an understanding of the special requirements posed by lens design problems. The simulated annealing algorithm for global optimization has attracted considerable attention from lens designers; existing variants of simulated annealing, however, lack the desired invariance under linear transformations of the coordinates or of the objective function, and typically require the specification of several hand-picked, problem-specific parameters. Success in applying these methods to lens design problems has therefore been limited. A new variant of simulated annealing-Adaptive Simulated Annealing (ASA)-is presented here in the general context of global optimization over continuous variables subject to constraints. In ASA, fixed temperature schedules and step generation routines have been replaced with heuristic-based algorithms that require minimal user-specified information and that are invariant under linear transformations of either the coordinates or the objective function. A parallel-processing implementation of ASA that provides increased efficiency is presented and applied to two standard problems for illustration and comparison. The application of ASA to lens design is examined. It is found that the lens design problem as typically posed-the canonical lens design problem-is not suited to efficient optimization with ASA. Modifications for optimization with ASA that raise efficiency-in particular, nonlinear inequality constraints and nonlinear coordinate transformations-are introduced. The effectiveness of ASA in lens design is then demonstrated through application to several standard lens design problems.