To deal with the situation with given traffic demand generated from origins and partial observed link flows, a bi-level programming model of estimation of the OD demand is proposed. The upper level was designed to minimize the difference between the assigned link flows and the observed link flows. To properly describe the route choosing behaviors of travelers, the user equilibrium traffic assignment model was adopted in the lower level model. A proper algorithm was designed for the upper level model by embedding an approximate FrankWolfe algorithm into the augmented Lagrange multiplier algorithm. To obtain the joint solution of the bi-level model, the algorithm of lower level model with given OD demands should be called repeatedly to obtain the assigned link flow. The numerical example verified the effectiveness and efficiency of the new model and algorithm. The research result not only extends the application setting of the existing theory of estimation of OD demands, but also provides new ideas and methods to formulate and analyze the related problems.