A flame image segmentation method was proposed based on reciprocal cross entropy threshold selection and bee colony optimization to improve the segmented accuracy. By using the minimum reciprocal cross entropy as the threshold selection criteria, the drawback of an undefined value at zero in Shannon entropy definition was avoided. At the same time, the 2D histogram oblique segmentation was taken to partition the object and background precisely to improve the anti-noise performance. By which, only one threshold instead of two thresholds needs to be searched, and the running time is reduced. In addition, the bee colony optimization was applied to acceleration of the process to find the optimal threshold to further improve the real-time performance of this algorithm and increase the algorithmic speed by 80%-140%. Finally, a large number of experiments on flame images were processed and then the experimental results were compared with the maximum Shannon entropy method based on 2D histogram oblique segmentation and the maximum reciprocal entropy method based on 2D histogram oblique segmentation and Niche Chaotic Mutation Particle Swarm Optimization (NCPSO). The obtained results show that the proposed method has obvious advantages in segmentation effects and has better anti-noise ability and real-time performance for flame images.