Pneumoconiosis has been the most harmful occupational disease for coal miners for decades and is difficult to cure completely. If this disease can be diagnosed and cured in time, it can be effectively controlled. Currently, the most reliable clinical diagnosis method is using X-ray or high-resolution computerized tomography (CT). Here, we propose a method to diagnose pneumoconiosis using wrist pulse signals. First, a pulse sensor is used to collect wrist pulse signals, then, preprocessing is conducted, and a single period of a pulse signal is separated. Second, a 13-dimensional feature is extracted in the time, frequency, and wavelet domains. Finally, a support vector machine, back propagation neural network, or random forest classifier are used to classify the pulse signals, two voting procedures are used and the diagnosis result can be achieved. The lowest recognition accuracy is 85.16%. The experimental results show that our 13-dimensional feature can be used as the main feature of pneumoconiosis diagnosis. The proposed method provides a more reliable auxiliary means for the diagnosis of coal mine pneumoconiosis.