An investigation into the turbo decoder is presented. It was carried out to determine whether the decoded bit sequence constitutes a (at least) local maximum in the likelihood between possible codewords in the composite code trellis. The answer was obtained experimentally using a modified iterative turbo decoder. Results show that the turbo decoder does not necessarily lead to a maximum likelihood sequence estimator being realised for the composite code. Moreover, finding a closer code in a Euclidean distance sense only degrades the bit error rate of the system.