Fingerprint-based indoor positioning estimates the users' locations in wireless local area network environments where satellite-based positioning methods cannot work properly. In this method, the location of a user is estimated by a pattern recognition algorithm (PRA). Traditionally, the training phase of PRA is conducted for and coordinates separately. However, the received signal strength from access points is a unique fingerprint for each measured point, not for and coordinates, independently. In this letter, we propose a novel PRA-based Gaussian process regression (GPR) method, named 2D-GPR, to jointly employ the and coordinates during the training phase. Experimental results show the superiority of 2D-GPR over conventional GPR (CGPR) and other competitors, especially in limited data samples. Also, the proposed method has a lower computation complexity compared with CGPR.