Optimum multiuser detection (OMD) for CDMA systems is an NP-complete combinatorial optimization problem. Fitness landscape has been proven to be very useful for understanding the behavior of combinatorial optimization algorithms and can help in predicting their performance. This paper analyzes the statistic properties of the fitness landscape of the OMD problem by performing autocorrelation analysis, fitness distance correlation test and epistasis measure. The analysis results explain why some random search algorithms are effective methods for OMD problem and give hints how to design more efficient randomized search heuristic algorithms for OMD.In direct-sequence code-division multiple access (DS-CDMA) communication systems, transmitters multiply each user's signal by a distinct code waveform. Detectors receive a signal composed of the sum of all active users' signals, which overlap in time and frequency. A particular user's signal is detected by correlating the entire received signal with that user's code waveform without regard for the other users, which inevitably yields multiple access interference (MAI) at the output of matched filter. MAI is the main factor limiting performance in CDMA systems. While optimum multiuser detection (OMD) [1] , which is based on the maximum likelihood sequence estimation rule, is the most promising technique for mitigating MAI, its computational complexity increases exponentially with the number of active users, which leads to its implementation impractical.From a combinatorial optimization viewpoint, the OMD is an NP-complete problem [2] . Randomized search heuristics (RSH) are effective methods for such kinds of problems, so many RSH based multiuser detectors have been studied and exhibit better performance than that of the other linear or nonlinear detectors. Earlier works on applying RSH to OMD problem can be found in [3][4][5][6][7] .The essence of optimum multiuser detection is to search for possible combinations of the users' entire transmitted bit sequence that maximizes the logarithm likelihood function (LLF) derived from the maximum likelihood sequence estimation rule [1] , which is called fitness function or objective function in the RSH multiuser detectors [3][4][5][6][7] . Comparing with so much emphasis on the implementation details and the performance analysis of these algo-