Bit-interleaved coded modulation with iterative decoding (BICM-ID) is a better scheme than BICM through applying iterative decoding algorithm at the receiver. It can get excellent performances both in AWGN and in Rayleigh fading channels. However, the iterative algorithm increases the complexity of the receiver and leads to much decoding delay. Thus, the stopping criterion, which can timely terminate the unnecessary decoding iterations, is put forward. In this paper, based on the cross-entropy (CE) concept, we propose a new method named LLR (log-likelihood-ratio) stopping criterion. Simulation results show that without any exponential computation, the proposed LLR scheme can take a better trade off between the average iteration numbers and the bit-error-rate (BER) performance degradation. Although with a little more iterative numbers than CE at high signal-to-noise ratio (SNR) sections, it obtains excellent BER performance as compensation. More importantly, besides the less iteration numbers, the LLR scheme performs even better than the fixed scheme at high SNR sections in terms of BER performance.