OBJECTIVES: The objective of this study was to compare estimates from survey and claims data sources for a low-volume procedure within a specialty population (specifically, the annual number of inpatient colonoscopies performed in the United States (US) pediatric population with private insurance) to better understand the considerations when choosing a data source. METHODS: A retrospective analysis of US health insurance claims and national survey data was performed using Thomson Reuters MarketScan® Commercial Claims and Encounters Database (MarketScan®), the National Hospital Discharge Survey (NHDS), and the National Healthcare Cost and Utilization Project Kids’ Inpatient Database (KID). Estimates among the privately-insured were obtained overall and by age (0 to 17 years), using ICD-9-CM procedure code 45.23 for calendar year 2007 in MarketScan® and NHDS and 2006 in KID (2007 data not available). RESULTS: The overall, annual estimate of inpatient, privately-insured pediatric colonoscopies was similar between NHDS and MarketScan®; however, NHDS estimates were not statistically reliable and differed dramatically from the prior year (overall estimate was 5.6 times higher than 2006). The overall estimate from MarketScan® was approximately 4.3 times higher than KID with estimates by age also higher; 2.8 times for age <1 to 4.2 times for age range 15-17. The KID database reported that approximately 50% of colonoscopies were performed in pediatric hospitals. CONCLUSIONS: The estimates were markedly different between MarketScan® and KID, and were not statistically reliable in NHDS. Low representation of pediatric hospitals that perform a large number of colonoscopies in this population possibly led to inconsistent or underestimated projections in NHDS and KID, respectively. Researchers should be informed about the frequency of the procedure of interest and representativeness of the population in the data when selecting a data source. Further research is warranted with other low-volume procedures performed in specialty populations to substantiate these findings.