We analyze and compare the maximum-likelihood (ML) decoder and the maximum a-posteriori (MAP) decoder for coding over a decode-and-forward relay network with orthogonal channels. Using a specific model for the errors introduced by the relay, we derive an optimal MAP decoder that operates on a super-trellis and estimates jointly the transmitted codeword and the error pattern introduced by the relay. Furthermore we present analytical expressions for the error rates for both ML and MAP decoding.