The invention relates to the field of financial stocks and particularly relates to a stock data processing method and device. The method comprises the steps of obtaining the time sequence data of stock, obtaining a weight factor q, obtaining a time scale delta, obtaining a probability measure Pi (delta) according to the time scale delta, obtaining a partition function x q (delta) according to the probability measure Pi (delta), judging whether the time scale delta selection is completed or not, obtaining double logarithmic curves ln xq(delta)-ln delta when the time scale delta selection is completed, obtaining the fitting slope tau (q) of the double logarithmic curves ln xq(delta)-ln delta in a scale-free network, carrying out Legendre transform on the fitting slope tau (q) and obtaining f (alpha), obtaining q+1 according to the weight factor q, judging whether q+1 is smaller than qmax or not, and obtaining a curve f (alpha)-alpha when the q+1 is not smaller than qmax and obtaining a multifractal spectrum and a spectral width through the curve f (alpha)-alpha. Through the above technical scheme, the change rule of the fluctuation before and after a stock crash can be grasped, and the interval of the stock crash is clear.