A new fusing facial feature recognition algorithm based on kernel Canonical Correlation Analysis (Kernel CCA) was proposed,for mapping image data into feature space and improving classifying accuracy.In our approach,we first map the image data through kernel function,then extract feature from the directions of rows and columns.Our algorithm simplifies the computation without decomposing the mapped matrix and gains the more discriminated features.The experiment results on OTCBVS V/IR face database of Ohio state university show that our algorithm gets better performance than other facial recognition method based on CCA with recognition accuracytate more than 90%.In addition,it also can get the excellent results with the illumination and expression variance.