In this paper, we propose a formulation of logistic regression for multiway (i.e. data where the same set of variables is collected at different occasions). More specifically, multiway logistic regression (MLR) constraints the coefficients of the logistic model to a tensorial structure that fits the natural structure of the data. Expected improvements of MLR compared with Logistic Regression are (i) better interpretability of the resulting model that allows studying separately the effects of the variables and the effects of modalities, and (ii) limit the number of coefficients to be estimated that decreases the computational burden and allows a better control of the overfitting issue. An aternating directions algorithm is proposed for MLR and the performances are evaluated on simulated data.