TransportTP is a two-phase transporter prediction and characterization system which may work on a genome scale. First, traditional homology methods are employed to predict novel transporters based on similarity to known classified proteins in the Transporter Classification Database (TCDB), during which pairwise and domain similarities are integrated. Second, machine learning methods are used to integrate a variety of features to refine the initial predictions, during which both non-homology evidence and homology evidence from other sources are integrated, such as transmembrane segments and the top-K nearest neighbors in TCDB, homologs in Pfam and Gene Ontology, and non-transporter homologs from Swissprot. All forms of evidence are converted into features of a refining classifier and rules to discriminate between true positives and false positives of the initial classifier are learned from some well-studied model organisms. A cross-validation using the yeast proteome for training and the proteomes of ten other organisms for testing, TransportTP achieved an equivalent recall and precision of 81.8%, based on TransportDB, a manually annotated transporter database for hundreds of organisms. In an independent test using the Arabidopsis proteome for training and four recently sequenced plant proteomes for testing, it achieved a recall of 74.6% and a precision of 73.4%, according to our manual curation. Comparing with our previous approach which only adopted BLAST search, HMM modeling and TMS filtering [ Li et al, Bioinformatics, 2008,server], the two-phase system increased more than 20% of recall with comparable recall to TransportDB. It also outperformed other alternative approaches, such as BLAST, BLAST plus HMM, or traditional Support Vector Machines (SVMs).