As an important methodology for optimally merging Earth observation information and geophysical model output information, data assimilation has played an important role in the area of Earth observation. At present, great progress has been made in the theoretical and methodological exploration and foundation of the operational land data assimilation system. Due to the complexity of research objectives, error problems are thought to be the bottleneck for improving the performance of data assimilation systems. Firstly, the research statuses of error problems of Land Data Assimilation Systems are reviewed. Based on the mathematical descriptions of land surface process model and measurement process, error sources and error characteristic are unifying defined. In a word, data assimilation systems include model errors, observation errors and the algorithm errors. Secondly, with respect to the sequential and variational data assimilation methods, error definitions and the related theoretical problems of those methods are briefly introduced with the emphasis on the error sources and the fundamental error parameterization methods. Moreover, from the perspective of error estimation, several novel methods for estimating model errors are reviewed from three parts: the model input error estimation, the model parameters error estimation and the model structure error estimations. As for the observation errors, the error sources can be divided with the observation algorithm errors, the representative errors and the instrument errors. Beside some exiting methods, there are no more effectively methods to deal with those kinds of error. Meanwhile, the difficulties for implementing all those methods are clarified. Thirdly, in order to reduce the errors for ensemble data assimilation systems, the common error parameterization methods, such as multiplicative inflation methods, additive inflation methods and the relax-to-prior methods, are employed. All these methods for dealing with model errors are meant to ameliorate the bias error in ensemble second moment. As far as the model bias is concerned, the state augmentation methods are discussed. A new scheme to obtain the optimal estimation of the state and model bias variable simultaneously is reviewed. Finally, the characteristic of all kinds of error estimation and processing methods and the surveys for the future implementation of all above methods in land data assimilation are given.