The National Meteorological Center (NMC) Regional Optimum-Interpolation (ROI) analysis is described. The ROI is the analysis component of the Regional Analysis and Forecast System (RAFS) and is specially designed to provide initial conditions for the Nested Grid Model (NGM), the forecast component of the RAFS. The ROI is an attempt to overcome weaknesses in the Limited-area Fine Mesh (LFM) and Global Optimum-Interpolation (GOI) analysis systems, to provide the NGM with more detailed and balanced analyses and to do so in a three-dimensional and dynamically consistent manner. Among the unique aspects are its hemispheric domain with regional-scale (150–200 km) horizontal resolution and improved vertical resolution (16 levels with the first 12 below 250 mb). The analysis is geostrophically coupled and is multivariate in geopotential and wind; it utilizes the same sigma coordinate as the NGM prediction model. All significant-level radiosonde data are now used, as well as many more of the surface observations. The first-guess fields are adjusted to an improved terrain representation and treated directly in the sigma coordinate. Single-level observations are combined in a simple “super observation” technique which averages nearby reports. The first-guess error correlation functions are more peaked in response to the increased resolution of the analysis. The observation error standard deviations have also been reduced to account for decreased errors of representativeness. The degree of geostrophic coupling, the relative error levels of mass versus wind data and the scale of the horizontal correlation function are varied during the analysis at the uppermost levels and at the levels nearest the surface. The observed surface temperature is now used in the analysis of heights. Finally, in all stages of the analysis, there is an increased dependence on observation quality codes. Verifications versus North American radiosonde data indicate a somewhat looser fit to the data than that of the LFM analyses, but the fit is tighter than the GOI. The fit of moisture data is better than either the LFM or the GOI analyses. Examples of analyses are shown with comparisons to the LFM analyses and first-guess fields. Although the regional analyses tend to underestimate wind speeds at 250 mb, they exhibit increased detail in the moisture and vorticity fields.