Evaluating the performance of stereo algorithms, both in terms of robustness and precision, is of critical importance as they become more amenable to practical applications. It is, however, a hard task because no unified testbed exists. In this paper, we use the algorithms that have been developed at INRIA in recent years to propose a number of methods that can be used to achieve this task and evaluate the appropriateness of algorithms for given applications. ∗Support for this research was partially provided by ESPRIT P2502 (VOILA), the CNES VAP contract, the EUREKA Prometheus Contract and a Defense Advanced Research Projects Agency contract (at SRI int.).