Aiming to implement the intelligent early warning of forest fire, a method based on color and texture features was proposed for forest fire smoke recognition. First, the color features were used to determine the smoke suspected area. Besides, the local binary pattern variance(LBPV) was utilized to extract the irregular feature of texture in the suspected area, and the LBP images was got. Wavelet transform was then used to extract the fuzzy, complex and correlative features from LBP images. At last, the fire smoke was identified by support vector machine(SVM). The result demonstrated that the method based on color and texture features has good recognition of forest fire smoke, which provides an effective solution for the study of forest fire smoke recognition.