During iterative decoding, a monotonic increase in the magnitude of the log-likelihood ratios is a sign of convergence to the maximum likelihood solution. When the log-likelihood ratios undergo frequent sign changes during decoding, the decoder generally fails to converge and the number of bit errors within a frame increases. In such cases, little gain is achieved by performing further iterations. Therefore, utilizing iteration termination methods is beneficial for minimizing the decoding latency, and reducing the operational complexity of turbo decoders. The stopping criteria that have been proposed until now are mainly based on the convergence of log-likelihood ratios of the constituent component decoders. In this paper, we propose a new criterion based on the prediction of the number of bit errors in a decoded frame. The iterations are stopped when the bit error rate of the turbo decoder falls below a specified threshold. This technique is found to be particularly effective in the waterfall region.