OBJECTIVES: Administrative claims data are commonly used to study CRC treatment patterns and outcomes.  We critically review existing algorithms for identifying incident CRC cases and disease stage at diagnosis within claims data. METHODS: A literature search (1989-2012) using the terms: cancer [ti] AND (administrative OR claims) AND (assess [ti] or assessment [ti] or predict [ti] or prediction [ti] or identify [ti] or identification [ti] or validate[ti] or validation[ti] or agreement[ti] or agree[ti]) identified 76 articles with 27 testing algorithms.  Of the 27, two provided algorithms for identifying incident CRC cases and two algorithms identified metastatic CRC at diagnosis.  We examine the positive predictive value (PPV) of each algorithm and suggest revisions for improving the PPV. RESULTS: Setoguchi evaluated four algorithms for identifying CRC patients.  The first and most restrictive required combinations of ICD-9 and treatment codes.  The second used two diagnoses within two months, the third combined the first and second, and the fourth required one diagnosis.  Ramsey used one or more ICD-9 codes, the same as Setoguchi algorithm 4.  The PPV ranged from 45% to 71% for algorithms 4 to 1, respectively.  Disease stage algorithms included one by Anaya for identifying liver metastases (PPV= 87%) and one by Nordstrom for identifying metastatic disease at any site (PPV=80%).  CONCLUSIONS: For identifying incident CRC patients, we suggest revising Setoguchi algorithm 1 to follow the well-tested Nattinger breast cancer algorithm with updates to surgical/chemotherapy codes.  We also recommend fewer criteria among older patients, as PPV is lower for patients less likely to receive aggressive care.  For disease stage, metastatic disease can be accurately identified with a small set of codes, but no studies examined algorithms for identifying CRC stages I-III. We recommend adapting the Smith and Shih algorithm for breast cancer disease staging to CRC while incorporating the metastatic codes reviewed here.