The control period is a key tuning knob of all existing discrete-time self-tuners. An algorithm for its systematic data-based choice is presented in the paper. It relies on the theory of Bayesian structure determination applied to a special class of control-rate dependent regression models. The models describe the entire measured-data history even if the inputs vary with another rate than that attributed to the identified model. At the same time, if the input signal changes with a specific rate the corresponding model reduces to the standard SISO regression. The proposed algorithm modifies, in accordance with the observed data, prior probabilities assigned to the compared control periods. In a single formula, it weights the average predictive ability of the model; the intersampling behaviour of the output; the number of unknown model parameters and the number of data; and, the uncertainty of the parameters. Promising properties of the algorithm are illustrated by simulation results.