This paper presents a sliding window estimation method for simultaneous localization and mapping (SLAM) based on the predictionerror method (PEM). The estimation problem considers landmarks as parameters while treating dynamics using state space models. The gradient needed for parameter estimation is computed recursively using an extended kalman filter. Results from experiments and simulations with a monocular camera and inertial sensors are presented and compared to batch PEM and nonlinear least-squares SLAM estimators. The presented method maintains good accuracy, and its parametrization is well-suited for online implementation, as it scales better with the size of the problem than batch methods.