In simultaneous estimation of p normal means, James and Stein (1961) introduced an estimator, called Stein-type estimator, and showed that it has uniformly smaller risk than the maximum likelihood estimator (MLE) if p ≧3 under the squared error loss. Since then many authors have discussed the Stein-type estimator under various situations. In this paper we discuss the cases where constraints of linear inequalities are given on p normal means and give Stein-type estimator for means. We also give Stein-type estimator for means of Hudson class of probability distributions when constraints of simpler linear inequalities are given on means.