Utilizing the pattern of magnetic field intensity (MFI) variation over a distance as a fingerprint for indoor positioning of mobile devices has gained increasingly wide attention. However, pattern matching with a large fingerprint is computationally expensive, and magnetometers are vulnerable to ferromagnetic perturbations and prone to have unknown magnetic offset (MO) in MFI measurements, which may result in large positioning error. In addition, the time-varying attitude of mobile devices and unreliable attitude determination with power consuming gyroscopes in mobile devices can render the 3-D MFI measurement useless for indoor positioning. In this paper, we propose a low power consuming and computationally efficient indoor positioning technique using the difference in 2-D (horizontal and vertical) MFI measurements that are collected over a distance, where 3-D accelerometers are used to find the vertical direction, and the MFI measurement difference is used to enhance robustness to unknown MO fixed in the body frame but time varying due to user dynamics in the navigation frame. The analysis of real data verifies that the proposed technique affords superior positioning performance relative to the conventional magnetic fingerprint techniques and achieves a similar performance to the conventional magnetic fingerprint techniques with MO-free measurements