In recent years, smartphones have become the most popular devices in our daily lives. The step count is an important information for developing services for smartphone users, e.g., indoor localization and health management. Most step counting solutions restrict that (i) the phone has to be fixed to the user and (ii) the user cannot use the phone naturally while walking. We can see that these restrictions are inconvenient for users. In this paper, we propose an adaptive step calculation algorithm, which can relieve the above restrictions and can count users' steps precisely. The proposed algorithm is composed by two phases. The first phase collects linear acceleration and gravity values from the smartphone's accelerometer. Then, this phase transforms the perceived linear acceleration values to parallel with horizontal plane and identifies possible start points of periodical regular fluctuations (of linear acceleration measurements). The second phase adopts the concept of correlation coefficients to identify whether the collected sensing measurements exhibit similar tendencies, and then calculates step counts. In this work, we implement the proposed method on the Android platform. The experiment results indicate that the proposed scheme can accurately divide gait changes and effectively identify steps.