The strength and vertical distribution of atmospheric turbulence is a key factor determining the performance of optical and infrared telescopes, with and without adaptive optics. Yet, this remains challenging to measure. We describe a new technique using a sequence of short-exposure images of a star field, obtained with a small telescope. Differential motion between all pairs of star images is used to compute the structure functions of longitudinal and transverse wavefront tilt for a range of angular separations. These are compared with theoretical predictions of simple turbulence models by means of a Markov-Chain Monte-Carlo optimization. The method is able to estimate the turbulence profile in the lower atmosphere, the total and free-atmosphere seeing, and the outer scale. We present results of Monte-Carlo simulations used to verify the technique, and show some examples using data from the second AST3 telescope at Dome A in Antarctica.