A shortcoming of the hypoplastic model for clays proposed by the first author is an incorrect prediction of the initial portion of the undrained stress path, particularly for tests on normally consolidated soils at isotropic stress states. A conceptually simple modification of this model, which overcomes this drawback, is proposed in the contribution. The modified model is applicable to both normally consolidated and overconsolidated soils and predicts the same swept-out-memory states (i.e., normal compression lines) as the original model. At anisotropic stress states and at higher overconsolidation ratios the modified model yields predictions similar to the original model.