For representing the relative location relationship of local features in the image space, local feature Spatial Correlation Kernel (SCK) is proposed for image object classification. The local features in the image are extracted and quantized, and the spatial location auto-correlations are calculated for vector-quantized local features, and then the histogram intersection is used to match spatial location auto-correlations of two images to obtain the local feature spatial correlation kernel. The proposed kernel makes good use of both the powerfully discriminative ability of local features and their spatial locations. Furthermore, SCK has a linear computation cost, satisfies the positive definite condition and could be used for kernel-based learning algorithms. The experiments performed on the public image database by embedding SCK into the support vector machine to classify the image objects demonstrate that SCK achieves the good time efficiency and the good classification performance.