Herein, we propose an improved dynamic threshold cloud detection algorithm (I-DTCDA) for visible infrared imaging radiometers (VIIRS) based on the multi-channel, wide coverage, and short revisit period features of a VIIRS. In addition, the algorithm is also based on the characteristics of the cloud distributions and variations in the visible and thermal infrared channels. We validated the accuracy of the cloud detection results using the remote sensing visual interpretation method. We compared our results with those using the universal dynamic threshold cloud detection algorithm (UDTCDA) and the VIIRS cloud mask (VCM) products. The results show that the proposed algorithm has average overall accuracy of 93% (Kappa=0.821) over different surface features. In particular, for the thin and broken clouds, the overall accuracy is significantly improved and the commission and omission errors are obviously reduced. The cloud detection results using the proposed algorithm are superior to those using UDTCDA and VCM.