Device-specific faults are very common for mobile software applications. To avoid such faults and guarantee the reliability and quality of mobile applications, sufficient testing is required on different mobile devices, which is expensive and time-consuming. This makes the task of the optimal selection of mobile devices for testing important and interesting from both practical and theoretical points of view. The suggested approach in this paper is based on combinatorial methods for coverage of each device characteristics. The initial results of the experimental investigation using comparisons with a random selection of devices are provided and show that the proposed approach is effective and promising.