Common burden tests have different statistical performance in genetic association studies of rare variants. Here, we compare the statistical performance of burden tests, such as CMC, WST, SUM and extension methods, using the computer-simulated datasets of rare variants with different parameters of sample sizes, linkage disequilibrium (LD), and different numbers of mixed non-associated variants. The simulation results showed that the type I error for all methods is near 0.05. When the rare variants had the same direction of effect, the higher LD and the less non-associated variants, the higher the power of these method, except the data adaptive SUM test. When the direction was different, the power was significantly reduced for all methods. The methods that consider the direction yielded larger statistical power than those methods without considering the effect direction, except the strong LD condition. And the larger the sample size, the larger the power. The statistical performance of burden tests is affected by a variety of factors, including the sample size, effect direction of variants, non-associated variants, and LD. Therefore, when choosing the method and setting the collection unit and weight, the prior biological information of genetic variation should be integrated to improve study efficiency.