Fingerprint localization is one of the most promising indoor positioning methods, and the fingerprint model based on the wireless signal strength is widely used due to its no-additional hardware cost and easy-to-spread characteristics. The selection of the fingerprint model is the key factor to the fingerprint positioning accuracy. Although the traditional fingerprint method by selecting the fingerprint collection points can reduce the computation, it contributes little to the accuracy of the positioning. In this paper, a fingerprint model based on principal component analysis is proposed. The new model accomplishes improvement in positioning accuracy as well as reduction in fingerprint calculation by selecting a set of "ingredients" with the largest impact on the accuracy to guide the positioning of fingerprint. Experimental results show that compared with the fingerprint algorithms based on Euclidean distance and nearest neighbor, the fingerprint algorithm based on principal component analysis improves average positioning accuracy to 2.7 m from 5.3 m and 3.9 m.