Abstract: Because of characteristic of fractal dimension which present quantitatively describing of complexity of a sample data series and remarkable advantage of support vector machine (SVM) in small sample classification and regression, fractal dimension of signal data series is adopted as eigenvectors, and a novel detection method based on fractal and SVM is presented.Two models based on SVM are constructed.One is about flatters of spot welding and the other is about defect of nugget size.A array of SVM is consist of these two models.The array is used to detect the two defects synchronously.It is shows that this new method fits for nondestructive detection of spot welding from analysis of experiment results.This array of SVM can detect the two defects of flatters and the little nugget size better in process of spot welding