Breakthrough detection is of significant importance for preventing back striking in electrical discharge machining (EDM) of film cooling holes. To address the challenge of accurately capturing the breakthrough moment caused by electrode wear in EDM drilling, a breakthrough detection method is proposed based on Bayesian online change point detection. By sampling the feed speed of machining spindle as a feature signal, a dynamically updating probabilistic statistical model is established to describe the machining state of small-hole EDM process. The occurrence of breakthrough is detected by identifying changes in the model parameters. Furthermore, by quantifying the probability of abrupt changes in the machining stage, this method reduces the impact of transient instabilities on breakthrough detection during actual processing. Compared to the commonly used sliding window method, the detection robustness is significantly enhanced. Repeated experiments with different exit angles (0° and 45°) validate the effectiveness of the Bayesian online breakthrough detection method.