Chord recognition task is to split up a piece of music into segments and assign each of them a chord label according to the analysis of the harmonic content. Making the chord recognition task automatically to process audio recordings will be of great help for music information retrieval. Most of the existing chord recognition systems proceed as follows. In the first step, a given audio recording is converted into a sequence of chroma features. In the second step, the feature sequence is passed into chord recognition module in which the features are assigned with chord labels. However, although much research has been done, there is little understanding of the effect of the different processing stages and of the various parameter settings on the final recognition result. In this talk, we present a typical automatic chord recognition system and analyze the influence of its different stages. In particular, we consider several types of chroma features as well as several chord recognition methods based on simple templates and more statistical involved pattern matching. As the main contribution, we conduct extensive experiments to evaluate the impact of different modules. We particularly investigate the role of the parameters from both the feature side and the recognizer side systematically to reveal how they influence the overall performance.