The decoding complexity of the layered Alamouti code with the maximum-likelihood (ML) receiver is high. Partial interference cancellation (PIC) group decoding is an attractive method to reduce the ML decoding complexity. However, different grouping schemes for the PIC group decoding may have different performances. In this letter, by fully taking advantage of the Alamouti code structure, we analytically calculate the power gains of the information symbols after the PIC for a grouping scheme, where the zero-forcing (ZF) operator is used for the PIC. We then propose an optimal grouping scheme for the PIC group decoding by proposing a criterion to choose the grouping schemes adaptively so that the power gains of the information symbols are maximized after the PIC.