A new approach to blood vessel boundary estimation is presented in this paper. By appropriately modeling the blood vessel as a dynamically evolving state vector, and by taking into account the Poisson statistics of the x-ray imaging noise, we arrive at a state-space system with a nonlinear measurement equation which includes non-Gaussian, nonadditive noise. Maximum a posteriori (MAP) smoothing equations are derived for the state vector describing the vessel, and the optimally smoothed state vector is found by a dynamic programming search. This method performs especially well in images with low signal-to-noise ratio and low sampling rate. The performance of the proposed method is demonstrated by the boundary estimates obtained by applying the algorithm to a simulated vessel and measurement data as well as to real vessel phantom measurement data at various SNRs.