Kinematic data for gait analysis consists of joint angle curves plotted against percentages of the gait cycle. A typical gait analysis entails repeated measurement of the kinematic data. We present an automatic and computationally inexpensive method to choose the most representative curve and detect outliers amongst repeated curves. The method is based on the notion of depth, where the deepest curve is the equivalent to the median for univariate data. The method applies to single kinematic variable or multi-kinematic variables such as the gait profile. It is sensitive to both shape and position of the curves. A comparison with an existing statistical method is presented as well as an example on one patient s data.