In statistical parsing, the probabilistic models are used to evaluate the possibility of each candidate parse tree, where the parse tree with the largest probability is deemed to be the final result of the parsing. Therefore, the core of statistical parsing is a probabilistic evaluation model. The main difference among the various probabilistic evaluation models lies in which types of features in the context are used to assign the probabilities to the parse trees. Various probabilistic evaluation models have been proposed in the field of statistical parsing, where different models use different feature types. How to evaluate a feature type's predictive power for the parsing tree? The paper proposes an information theory based feature type analysis model. Using the method, we can quantitatively analyze the power of different feature types for syntactic structure prediction from the viewpoint of information theory. The basic idea is that we use entropy and conditional entropy to measure whether a feature type grasps some of the information for syntactic structure prediction. If the average uncertainty of the syntactic structures declines apparently, the feature type is deemed to have grasped some intrinsic linguistic information in the context that has close relation to the syntactic structure. Using Penn Treebank as training and testing set, our experiment quantitatively analyze the different feature types' predictive power for syntactic structure predictive power for syntactic structure prediction in a systematic way and draws a series of conclusions which reflect the predictive power of different feature types and feature type combination for syntactic parsing.