Ten, three year old, fodder tree species were evaluated at four on-station and three on-farm sites in Nepal. Ficus semicordata (Buchattam. ex Sm.) growth was found to be significantly higher than the rest in diameter and dry foliage weight values. Species were significantly different in height, diameter, and foliage and wood growth. Sites were significantly different in total height growth only. On-farm species evaluation indicated that A. lakoocha and F. semicordata had significantly higher growth. Allometric regression equations were developed to predict foliage, total wood, and total biomass yield of F. semicordata, and B. variegata. Individual-tree models were developed. For B. variegata, diameter at 50 cm. and for F. semicordata, crown diameter and height gave the best fitted equations. Regression equations for three sites did not differ significantly. Therefore, data were pooled and a common model was estimated for each species. In on-farm regression models, height and crown diameter were the best predictors for F. semicordata and dbh gave the best fit for B. variegata. The models for the two species were used to construct regional fodder and fuelwood biomass tables. An improved crop-livestock-fodder agroforestry system was designed for a village in Nepal. Linear programming was used to demonstrate the use of a tool to optimize land allocation maximizing net returns while satisfying the supply of minimum needs of food, fodder, and fuelwood. The optimal solution indicated that, by improving the returns to labor and by applying more compost, the village should be able to increase the annual net farm returns from NRs. 2.94 million to NRs. 3.85 million. The food, fodder and fuelwood production levels were shown to increase by 17%, 130%, and 537% respectively. The labor and compost requirements were up by 138% and 59% respectively, over the five year period. The soil loss through run-off was estimated to decrease by about 15% over the same period.