In this work, we proposed a novel method, Evolutionary Radial Basis Function Network (ERBFN), for classification of cancer types with microarray gene expression data. Evolutionary Radial Basis Function Network is a significant improvement over ordinary Radial Basis Function Network. Starting with traditional clustering algorithm, ERBFN optimized the hidden layer of Radial Function Network, and used supervised learning strategy to fine-tune the network connection weights. This method has been successfully applied to classification of real-world cancer data. Our assessment has revealed that the accuracy of ERBFN is comparable to that of support vector machine based classification.