In four communities in the Peruvian Andes, 56 farmers were interviewed every three months over a period of one year. Information linked to milk and cattle production such as activities, inputs (labour, means of production, capital) and outputs (milk, cheese, animals) were recorded using a closed-ended questionnaire. The communities were divided into two groups with low (LC) and high (HC) level of dependence on income from milk and animal sales. The survey results showed that cattle production on the LC farms was based on less land and a smaller herd (3.32ha/farm, 1.06 lactating cows) than on HC farms (10.28ha/farm, 4.19 lactating cows). The data from the survey and the results of the nutritional analyses of 74 feed samples were introduced into a model that applied linear programming techniques in order to estimate the farm household income under the current production systems and evaluate the economic impact of improved forage varieties for hay production. Furthermore, the economic viability of other changes in fodder and herd management was tested. Both groups were characterised by a dual-purpose system generating a gross income from the sale of both, milk and live animals in the amount of -21 (LC) and +1057 US$/farm and year (HC). Due to higher production costs for forages and better access to markets, LC communities were characterised by an integrated crop–livestock system whereas in the HC group income was mainly based on livestock. Introduction of improved and fertilized barley for hay production, was estimated to increase the annual farm income to 127 and 1257 US$ for LC and HC, respectively. This increase was accompanied by an increment of the animal number. Maintaining the animal number but increasing the milk production/cow by feeding additional forage was a less profitable option generating 50 and 1221 US$ of income per farm and year for LC and HC, respectively. The production of hay was limited by high costs (external labour) in LC communities and the restricted availability of family labour in the HC group. A scenario based on the use of improved cow genotypes led to the highest estimated annual farm income for HC communities (1280 US$) but was less favourable for LC. The modelling results showed that the best development strategy depends on various factors such as production costs, access to the markets and to irrigation and availability of different feed resources.