Based on the dynamic surface control(DSC) technique, and by using the approximation capability of radial basis function(RBF) neural networks, an output feedback adaptive tracking control scheme is proposed for a class of nonlinear systems with input and state unmodeled dynamics. K-filters are designed to estimate the unmeasured states. By using the properties of the Nussbaum type function, the problem of the unknown high frequency gain sign is effectively solved. A normalization signal is designed to restrict the input unmodeled dynamics, and the disturbance caused by it is effectively suppressed. By theoretical analysis, it is shown that all the signals in the closed-loop system are semi-globally uniformly ultimately bounded.