The demand for higher data rates, capacity and better Quality of Service (QoS) is constantly growing; therefore, there is a pressing need for efficient use of wireless network resources. In this context, the application of Cognitive Radio (CR) principle has emerged as a solution to the problem of spectrum scarcity. In this respect, this paper presents an algorithm that enables multi-band spectrum selection for wireless applications under interference variations in the available spectrum blocks. The algorithm is formulated as a Partially Observable Markov Decision Process (POMDP) targeting the maximization of a reward function that captures how suitable each allocated spectrum block is for the data transmission in each radio link depending on the application bit rate requirements. ; Peer Reviewed ; Postprint (published version)