Abstract
The paper considers sequential estimation of the difference of two -variate normal means from independent populations when the variance covariance matrix from each population is known up to a scalar multiple and are not necessarily equal. The sampling scheme is a multivariate extension of the stopping rule proposed by Ghosh and
Mukhopadhyay (1980). A class of James Stein estimators that dominates the difference of sample mean vectors is developed and asymptotic risk expansions are also provided.