A challenge in echocardiography is segmentation of the heart muscle. The main problem is a similar echogenicity of blood and heart muscle for some regions. Therefore, segmentation based on differences in temporal decorrelation between blood and tissue might be a promising tool. In 3D Doppler mode, a sequence of eight consecutive flow lines is recorded for each B-mode line. Each set of flow lines is highly correlated, because of the high pulse repetition frequency. Due to blood flow, however, decreasing correlations are found with increasing flow rate. These decorrelation values were used as temporal constraints for a deformable simplex mesh model. A simplex mesh model is used with an internal energy term, representing topological information about the mesh distribution and an external temporal decorrelation term. The model was initialized at a seed point in a region of interest. Unlike most deformable models, the mesh did not evolve on the basis of gradient information, but by connectedness obtained from a local fuzzy relation called 'affinity'. Affinity takes into account the degree of adjacency of grid points as well as the similarity of the region values. This model was validated in a phantom study that simulated parabolic flow in an artificial artery at different flow velocities and preliminary evaluated for segmentation of 3D cardiac flow data of the left ventricle. The deformable simplex mesh model successfully segmented the artificial blood vessel based upon flow constraints. For low velocities the vessel volume was underestimated. With increasing flow, this underestimation decreased and converged to the real volume. Preliminary inpatient evaluation revealed appropriate segmentation of consecutive blood volumes. In conclusion, deformable simplex mesh models with temporal signal decorrelation constraints might be a useful approach for 3D segmentation in ultrasound.