In the field of isolated handwritten word recognition, the development of verification systems that optimize the trade-off between performance and reliability is still an active research topic. To minimize the recognition errors, usually, a verification system is used to accept or reject the hypotheses produced by an existing recognition system. In this paper, a novel verification architecture is presented. In essence, the recognition hypotheses, re-scored by a set of the support vector machines, are validated by a verification mechanism based on multiple rejection thresholds. In order to tune these (class-dependent) rejection thresholds, an algorithm based on dynamic programming is proposed which focus on maximizing the recognition rate for a given prefixed error rate.