As one of the most important reversible protein post-translation modification types, ubiquitination plays a significant role in the regulations of many biological processes, such as cell division, signal transduction, apoptosis and immune response. Protein ubiquitination usually occurs when an ubiquitin molecule is attached to a lysine on a target protein, which is also known as “lysine ubiquitination”. In order to investigate the molecular mechanisms of ubiquitination-related biological processes, the crucial first step is the identification of ubiquitination sites. However, conventional experimental methods in detecting ubiquitination sites are often time-consuming and a large amount of ubiquitination sites remain unidentified. In this study, an ubiquitination site prediction method for Arabidopsis thaliana was developed using support vector machine (SVM). We collected 3009 experimentally validated ubiquitination sites on 1607 proteins in A. thaliana to construct the training set. Three feature encoding schemes were used to characterize the sequence patterns around ubiquitination sites, including AAC, Binary and CKSAAP. The maximum relevance and minimum redundancy (mRMR) feature selection method was employed to reduce the dimensionality of input features. Five-fold cross-validation and independent tests were used to evaluate the performance of the established models. As a result, the combination of AAC and CKSAAP encoding schemes yielded the best performance with the accuracy and AUC of 81.35% and 0.868 in the independent test. We also generated an online predictor termed as AraUbiSite, which is freely accessible at: http://systbio.cau.edu.cn/araubisite .