Decision-making approaches, the factors affecting them, the role of the decision-making units and the actors involved in farming decisions were studied in 91 Costa Rican farmers using a classification procedure. Farmers were asked to classify 18 farming decisions (systematically selected from a pool of 100) into a series of categories of decision-making units (Alone, Family, Shared, Delegated) and actors (person/people involved). Frequency tables and multivariate analyses were used to analyse the data. A canonical correlation analysis (CCA) was performed to find out simple and canonical correlations between farmers'/farms' characteristics and the decision-making approaches. Factor analysis (FA) combined with a Cluster analysis (CLA) was used in order to: (1) define groups of farmers with similar decision-making approaches; and (2) define groups of decisions made by similar decision-making units. Logistic Regression was used to study the influence of the intrinsic characteristics of the decisions and the decision-making units involved. A Multiple Correspondence Analysis was used to graphically represent the relationships between shared and delegated decisions and the actors involved. The CCA showed that the level of dedication to farming, the educational level and the size of the farm were the characteristics more strongly influencing the decision-making approaches. Monopolisation of decisions by one decision-maker was the most frequent decision-making approach in the population since half of the decisions were made by the farmer alone. However, other approaches in which other decision-making units, mostly the family, had an important role were also frequently found in the population. FA and CLA found very well defined groups of decisions in terms of decision-making units involved. The logistic regression suggested that the intrinsic characteristics of the farming decisions influenced the level of involvement of the different decision-making units. Operational decisions tended to be more delegated to farm staff and family members, while technical decisions were mostly shared mainly with technical advisors and family members. Implications of these findings on research and extension activities are also discussed.