Aiming at problem that segmentation threshold value of a ferrographic image is difficult to select in oil monitoring, a self-adaptive ferrographic image segmentation algorithm based on difference quotient was introduced. Firstly, the ferrographic abrasive particle image was converted into three-dimensional grey histogram and then a slice analysis was made on it; then, by introducing Newton interpolation polynomial, the pixel number obtained from different slices was took as interpolating point of slice grayscale-frequency curve; the first kind of acceptable function and the second kind of acceptable function were established based on difference quotient, and two kinds of errors were identified by combination of experimental data. The minimum gray value which simultaneously satisfied the two kinds of errors was selected as segmentation threshold value. Finally, segmentation experiments on different types of ferrographic images and ferrographic images with Gaussian noise and salt & pepper noise were conducted to compare the performance of proposed algorithm and three classical algorithms including iterated thresholding method, Otsu algorithm and maximum entropy. The experimental result indicates that the proposed algorithm is rarely interfered by noise and its average false positive rate and average omission rate is overall superior to other three algorithms. Through conducting feature extraction on ferrographic image and identification by support vector machine, it can be found that the proposed method has the highest identification accuracy rate on three faulted abrasive particles, which reaches 82.86%. Although there are no obvious advantages on operation time, but the method has optimal comprehensive property and can meet the requirement for making a self-adaptive segmentation on ferrographic image in the process of oil monitoring.