This paper formulates a state observer based nonlinear model predictive control (MPC) technique using successive linearization. Based on local linear approximations of state/measurement equations computed at each sample time, a recursive state estimator providing the minimum-variance state estimates (known as the 'extended Kalman filter (EKF)'') is derived. The same local linear approximation of the state equation is used to develop an optimal prediction equation for the future states. The prediction equation is made linear with respect to the undecided control input moves by making linear approximations dual to those made for the EKF. As a result of these approximations, increase in the computational demand over linear MPC is quite mild. The prediction equation can be computed via noniterative nonlinear integration. Minimization of the weighted 2-norm of the tracking errors with various constraints can be solved via quadratic programming. Connections with previously published successive linearization based approaches of nonlinear quadratic dynamic matrix control are made. Under restrictive assumptions on the external disturbances and measurement noise, the proposed algorithm reduces to these techniques. Potential benefits, hazards and shortcomings of the proposed technique are pointed out using a control problem arising in a paper machine.