Quantum annealers are analog quantum computers design to solve hard combinatorial problems called Quadratic Unconstrained Binary Optimization (QUBO). For an input QUBO, a quantum annealer output at random several configurations of binary variables. In this talk we introduce a method to learn and characterize the probability distribution of outputs of quantum annealers. We present experimental reconstructions for the specific D-Wave system architecture and relate the probability distribution to the chip architecture.