This paper presents a sampling method of multidimensional correlated wind speed that combines with improved correlation matrix Monte Carlo simulation method and Latin hypercube sampling method. The proposed method has been applied to composite generation and transmission system reliability evaluation with multiple wind farms. The tests using actual wind speed data verify that the proposed model for wind speed correlation can capture the statistical properties of actual wind speeds, including probability distributions, basic statistics and correlations. The results from the application of the proposed method to the MRTS reliability test system demonstrate that this method is appropriate for reliability evaluation of composite generation and transmission system containing multiple wind farms with correlations. And according to some examples, the effects of wind speed correlation, position of wind farm and wind power penetration level on composite generation and transmission system reliability including wind farms are analyzed.