Today's communication system design heavily depends on computer simulation for performance evaluation. The burgeoning ultra-reliable communication systems, however, pose a significant simulation challenge as such systems operate at very low packet error rates (PERs) whereas the required simulation time of the conventional Monte Carlo (MC) method is many times the inverse of the PER. Various importance sampling-type techniques have been developed for more efficient simulation of channelcoded transmission, but they typically rely on exploiting code weaknesses (for the generation of errorcausing noise samples) and most works on soft-decision decoding only treat binary signaling. In this paper, we propose to use a function, termed the noise gauging function (NGF), that roughly measures the error-causing propensity of noise samples and we present a way to adaptively optimize the noise sampling under such a function for simulation efficiency. Both binary and nonbinary signalings are considered. And the proposed technique does not require detailed knowledge of the code weaknesses, although some high-level understanding of the code properties can benefit the design of efficient NGFs. We investigate the application of the proposed technique to several common channel codes. Numerical results indicate an approximately 10-to 1,000-fold speedup versus MC.INDEX TERMS BCH codes, channel coding, convolutional codes, fast simulation, importance sampling, Monte Carlo, packet error rate, polar codes, ultra reliable and low latency communication (URLLC).