Recently, wearable devices on human activity sensing have become increasingly popular. In this paper, we propose a new method for walking and running speed estimation. A wearable device based on a tri-axial accelerometer is fixed on the chest to record the acceleration data of human body. Using the acceleration data, we design a fuzzy inference system (FIS) to determine whether the subject is motionless or locomotor. If the subject is walking or running, artificial neural networks (ANNs) are then used to estimate the speed. To validate the performance of the proposed method, we test accelerometer data collected from 3 subjects walking and running on the treadmill at different speed. The result shows good agreement between actual and estimated speed, the average accuracy is 97.78% for walking and 97.36% for running.