In this paper, we introduce an Artificial Neural Network (ANN) classifier for human activity recognition. The proposed system is divided to three stages; first we do segmentation for raw data collected for two data sets WHARF and UCI-HAR to segment length 128 and 256 with 50% overlap for WHARF, and 128, 256 and 512 with 50% overlap for UCI_HAR. Second, for each segment, a set of time and frequency domain features are extracted and delivered to the ANN classifier. From a practical point of view, activity classification based on segments of data compared to the use of whole raw data is more suitable and enables faster classification process, especially for short activities. The proposed system is tested against other classifiers such as support vector machines (SVM), naive Bayes, and k-nearest neighbor (KNN) where ANN gives the best recognition rate. For WHARF dataset, the average accuracy is 68% for segment length 128 and 80% for 256 segment length. On the other hand, employing accelerometer data only in UCI-HAR dataset, the average accuracy is 93.9% for segment length 128, 94.6% for segment length 256 and 96.3% for segment length 512. While using gyroscope data in the same dataset results in an average accuracy of 77.96% for segment length 128, 83.75% for segment length 256 and 88.96% for segment length 512.