Sharp rises and falls in stock prices have become increasingly frequent in recent years. Stock market crashes bring great risks to the stability of the securities markets. Using recurrence plot theory and a heuristic segmentation algorithm for detecting abrupt changes in nonlinear time series, this study investigates the problem of detecting abrupt endogenous structural changes before a stock market crash. Based on an analysis of crash events in 12 developed and 10 emerging countries and regions, the authors find the following: (1) The market laminar flow (LAM) value will fall greatly before a stock market crash; (2) the LAM sequence of the US stock market during the 2008 financial crisis presents a fractal-like self-similar structure, and blank bands appears in the recurrence plot, indicating a phase transition in the LAM sequence before the crash; and. (3) using a heuristic segmentation algorithm to detect abrupt changes in nonlinear time series, this study finds that before a crash, the endogenous structure of the market continuously experiences abnormal abrupt changes, and abnormal abrupt change time.