A novel approach to the topic of analog-to-digital converter (ADC) characterization is proposed. The key idea is to describe the behaviour of the device via a suitable conditional probability function, estimated through a modified version of the popular histogram test. Any traditional figure of merit for ADCs can be accurately evaluated from the proposed probabilistic characterization. Besides, this allows one to optimize the ADC overall performance, determining the best allocation of the output reconstruction levels. The parameters of the modified histogram test are determined as a function of the desired accuracy. Finally, computer simulations illustrate the performance of the method