In this paper, three line fitting methods using k-means algorithm are proposed. The result of the early work by Phillips and Rosenfeld is dependent on the selection of initial partition, and the calculation of the principal axis of the data points needs many computations. To eliminate these shortcomings, we first propose an algorithm using an explicit initial partition according to the local property of distance deviation. This algorithm can obtain consistent and better results. Second, instead of computing the principal axis, a faster algorithm using a simpler line fitting method is presented. To determine the number of lines which approximate the given points automatically, another method using a useful criterion is proposed. The experimental results show that the proposed methods are feasible.