The structure of Nudaurelia capensis omega virus (NomegaV), a single-stranded RNA virus, was determined to 2.8 A resolution. Triclinic crystals (a = 413.6, b = 410.2, c = 419.7 A, alpha = 59.13, beta = 58.9, gamma = 64.0 degrees ) diffracted X-rays beyond 2.7 A resolution. The unit cell contained one icosahedral virus particle, providing 60-fold non-crystallographic symmetry (n.c.s.) and structural redundancy. The particle orientation in the unit cell was determined by self-rotation function analyses. Initial phases to 18 A resolution were derived from a hollow spherical model of 192 A outer radius and 139 A inner radius, filled with uniform electron density. Radii of the model were determined by maximizing the correlation of the model-based calculated data with the low-resolution X-ray diffraction and solution-scattering data. Phases were refined by 60-fold non-crystallographic electron-density averaging and extended in small steps to a resolution of 5 A. The phases obtained represented a mixture of four different phase sets, each consistent with the icosahedral symmetry constraints. The resulting electron density was not interpretable. A difference Fourier map computed with the native and an isomorphous heavy-atom derivative data sets and phases refined by real-space averaging was interpretable only if data within the 10 A resolution shell were used. Maps calculated with data significantly higher than 10 A resolution failed to display a constellation of heavy-atom sites consistent with the T = 4 icosahedral symmetry. Attempts to extend the phases beyond 10 A resolution, starting with either phases based on a model or single isomorphous replacement, were unsuccessful. Successful phase extension was achieved by computing the phases for the higher resolution reflections from a partial atomic model (poly gly) built into the averaged 10 A electron-density map. Phases from this model served as the starting point for n.c.s. phase refinement and extension to slightly higher resolution. The atomic model was improved at each extension interval and these phases were used for the subsequent phase calculation and extension. The entire polypeptide backbone corresponding to the NomegaV structure was built into the map at 4 A. The same procedure for phase refinement was used to extend the phases to 2.8 A in small increments of resolution. The overall molecular averaging R factor and correlation coefficient at 2.8 A resolution were 18.4% and 0.87, respectively.