Many software systems in financial firms are message- based server-side systems that perform important and reasonably complex business functions. The regression testing of these systems is currently done through a ‘log- replay’ mechanism, wherein the messages processed by the system in production are replayed through the new version to be tested, and the results compared (either against the production system or against expected results). As the production volumes are large and continue to grow rapidly, this approach leads to many problems such asscalability, resource crunch, reconciliation, redundancy, production impact.In view of these problems we propose an alternative approach to Log Replay Testing based on statistical sampling. We form an analogy of this problem with a variant of another problem called Coupon Collector Problem and use the solution to this problem with business logic appended to it.The sampling based approach that is proposed will essentially consist of two steps:1.???Extracting the set of message trees from the message log.2.???Sampling the set of extracted trees with the twin objectives of-a)Having a high degree of confidence in selecting all kinds of message trees andb)Selecting a small percentage of the set of trees.The proposed approach continues to use the production log as the basis for regression testing. However, it deviates from the current approach in not using the entire production log but only a subset or sample of it. If such a sampling can be done, then it is obvious that the number of test cases that will be executed will only be a small fraction of the number currently being executed, while still retaining a high degree of confidence in the quality of the selected sample.