This paper introduces a method for speculative execution, sentinel scheduling with recovery blocks, which is an improvement upon sentinel scheduling. This method is compared against speculative models that have been proposed previously. The hardware and compiler support needed for sentinel scheduling with recovery blocks is fully described. Quantitative results of speedup and code growth are presented for 14 benchmark programs. This method was fully implemented with the IMPACT compiler. Sentinel scheduling with recovery blocks is shown to have a low hardware cost, while achieving good performance results.