Two methods, namely combined plot analysis and integration of variograms, have been applied to investigate and compare plot-to-plot yields of intercropped millet (mean yield of about 1.9 t ha−1) and maize (mean yield 4.25 t ha−1) on terraces in Nepal. Combining plots diminishing the residual variance of millet from 0.52 (t/ha)2 of the original 1 m×1 m units to 0.08 (t/ha)2 for 4 m×4 m plots, the largest for which reliable estimates could be derived. The residual variance of maize declined from 4.91 (t/ha)2 of the original units to 0.61 (t/ha)2 for 4 m×4 m plots. Orientation was immaterial. The variogram for millet was isotropic and bounded with a range of approximately 4 m. Within-plot variance calculated from the variogram increased, and experimental error decreased, as the size of plots increased to this value, beyond which there was little gain. The variogram for maize was unbounded, and so gave no ‘natural' upper limit for plot size, that is, over the range studied, up to 6 m×6 m. It was also anisotropic with larger gradient in the direction from front to back of the terrace than along the contour. The reduction in residual variance was greatest for narrow plots elongated in this direction. Formulae are provided from which to calculate from the variogram approximate residual variances for experiments of given area and size, shape, and orientation of plots. Its application gave similar results to those from the combined plot analysis. Overall, the results from the two methods were consistent.