Group sequential design is widely used in today’s phase II/III clinical trials where testing multiple endpoints is quite often performed. In such tests, a basic requirement is to control the family-wise error rate at a given nominal level. The design is determined by a set of testing statistic and stopping boundaries (rules). Existing methods compute the stopping boundaries use Normal approximations, which work well when the true underlying data distribution is approximately Normal, but with small sample sizes the Normal approximation may not be valid. In an attempt to overcome these difficulties, we propose a robust method to compute the stopping boundaries in which it is assumed only that the data distributions are symmetric about their means. The null and alternative distributions are then constructed via the empirical distribution as well as the stopping boundaries for the given nominal level. Powers for the test statistics are obtained by bootstrap simulation, which is always valid for any sample size, and correlations between test statistics are automatically taken care of. Simulation examples are given to illustrate the proposed method.