A digital raster map called a 'model forest' was generated using sample plot data from the Nepal national inventory of 1991 (Kapiolvastu and Ruoandehi districts), aerial photographs of 1989- 90, and Landsat Thematic Mapper (TM) satellite imagery of the area for 1990. The model forest was a map representing the spatial variation of stem volume, and was used to to simulate different cluster sampling techniques for forest inventory in southern Nepal. In the present plot (cluster) design used, each plot is made up of 3 circular subplots, with a distance from the cluster centre to the plot centre of 25 m. Designs were simulated with a varying number of (square) plots per cluster, varying spacing between plots, and with systematic or random sampling. The design of clusters was determined so that it was possible to measure the cluster in 1 day. The different cluster designs were compared using the precision (sampling error) and efficiency (time and cost) of the technique as criteria. The optimal distance between the plots within the cluster was much longer than in the present clusters (25 m), in which the 3-plot cluster provided little additional information in comparison with a one- plot sample unit. The smallest sampling error (6.8 m³/ha) was obtained using a systematic triangular shaped 3-plot cluster sampling design with the plots spaced at 275 m from the cluster centre. When efficiency was also considered the design based on a systematic 8-plot square cluster with the plots spaced at 325 m was good, although less accurate (sampling error 10.5 m³/ha); only some additional clusters would have to be measured to increase accuracy.