Abstract Background Health researchers have increasingly adopted the use of geographic information systems (GIS) for analyzing environments in which people live and how those environments affect health. One aspect of this research that is often overlooked is the quality and detail of the road data and whether or not it is appropriate for the scale of analysis. Many readily available road datasets, both public domain and commercial, contain positional errors or generalizations that may not be compatible with highly accurate geospatial locations. This study examined the accuracy, completeness, and currency of four readily available public and commercial sources for road data (North Carolina Department of Transportation, StreetMap Pro, TIGER/Line 2000, TIGER/Line 2007) relative to a custom road dataset which we developed and used for comparison. Methods and Results A custom road network dataset was developed to examine associations between health behaviors and the environment among pregnant and postpartum women living in central North Carolina in the United States. Three analytical measures were developed to assess the comparative accuracy and utility of four publicly and commercially available road datasets and the custom dataset in relation to participants' residential locations over three time periods. The exclusion of road segments and positional errors in the four comparison road datasets resulted in between 5.9% and 64.4% of respondents lying farther than 15.24 meters from their nearest road, the distance of the threshold set by the project to facilitate spatial analysis. Agreement, using a Pearson's correlation coefficient, between the customized road dataset and the four comparison road datasets ranged from 0.01 to 0.82. Conclusion This study demonstrates the importance of examining available road datasets and assessing their completeness, accuracy, and currency for their particular study area. This paper serves as an example for assessing the feasibility of readily available commercial or public road datasets, and outlines the steps by which an improved custom dataset for a study area can be developed.