This paper proposes a practical method for estimating human walking speed using accelerometer data. A portable device based on accelerometer was developed to objectively record human walking signals. An Artificial Neural Network was developed to estimate the average speed of walking. We extracted six parameters, namely step number, subject’s height, root mean square and difference between the maximum and minimum of vertical and frontal accelerometer signals as inputs for artificial neural network. To validate the performance of the proposed method, we tested accelerometer data collected from 35 subjects walking under free-living conditions. The experiments shows that the average accuracy of speed estimate is 96.96% which better than previous works. Based on the results of these experiments, is concluded that a tri-accelerometer is a promising tool for the accurate assessment of walking speed.