Our knowledge of the proteome of plant peroxisomes is far from being complete, and the functional complexity and plasticity of this cell organelle are amazingly high particularly in plants, as exemplified by the model species Arabidopsis thaliana. Plant-specific peroxisome functions that have been uncovered only recently include, for instance, the participation of peroxisomes in phylloquinone and biotin biosynthesis. Experimental proteome studies have been proved very successful in defining the proteome of Arabidopsis peroxisomes but this approach also faces significant challenges and limitations. Complementary to experimental approaches, computational methods have emerged as important powerful tools to define the proteome of soluble matrix proteins of plant peroxisomes. Compared to other cell organelles such as mitochondria, plastids and the ER, the simultaneous operation of two major import pathways for soluble proteins in peroxisomes is rather atypical. Novel machine learning prediction approaches have been developed for peroxisome targeting signals type 1 (PTS1) and revealed high sensitivity and specificity, as validated by in vivo subcellular targeting analyses in diverse transient plant expression systems. Accordingly, the algorithms allow the correct prediction of many novel peroxisome-targeted proteins from plant genome sequences and the discovery of additional organelle functions. In contrast, the prediction of PTS2 proteins largely remains restricted to genome searches by conserved patterns contrary to more advanced machine learning methods. Here, we summarize and discuss the capabilities and accuracies of available prediction algorithms for PTS1 and PTS2 carrying proteins.