This paper derives and analyses a recursive algorithm for maximum a posteriori (MAP) state estimation of partially observed bilinear systems. The recursive algorithm is based on cross-coupling two Kalman filters, one for each component of the bilinear system