The measurement of joint angles is of profound importance in analysis of human gait. However, currently these measurements can be acquired only with the use of dedicated measuring devices such as a goniometer or in a gait lab with a motion capture system. The low-cost and portability of recent IMU technology makes them ideal for continuous monitoring of kinematic data. In this research we present an algorithm for estimating knee angles based on IMU data with the use of artificial neural networks. Two IMUs with tri-axis accelerometers and gyroscopes were used above and below the knee under investigation. Simultaneously, an electro-goniometer was used to measure the angle, acquired at 64 Hz. A feed-forward ANN with one hidden layer was fed IMU data from 25 of the acquired steps as inputs and trained against the goniometer angles. The ANN was evaluated on 25 trials for a single subject. The estimated knee angle exhibited strong correlation and minimal error compared to the actual angle.