Abstract: Panel count data are commonly encountered in follow-up studies such as clinical trials and sociological studies. Models for this type of data usually assume the underlying recurrent event process is independent of the observation process. However, the independence assumption cannot always be guaranteed in practical applications, especially when they both are related to the covariates. In this paper, we develop a semiparametric additive model weighted by propensity score for analyzing panel count data. By introducing the propensity score into the model, the confounding bias of parameter estimation caused by the dependent observation process may be reduced. In particular, we combined the inverse probabilityweighted estimation equations to estimate the parameters and further derived the asymptotic properties of the estimator. Numerical simulations were conducted to evaluate the proposed procedures. Finally, we applied the methodologies to analyze a set of skin cancer data.