Abstract Simple forcasting models are presented for farm tractor demand in Italy, France and the United States. The models tested are both univariate (ARIMA) and multivariate. The multivariate ones are constructed according to a demand function for annual tractor purchases and not according to a demand function for the tractor stock (or tractor fleet). The main independent variables for the multivariate model of the tractor demand (Yt) are the real agricultural added value or income of the previous year (X1,t-1), the real price index of tractors (X2,t), the tractor stock of the previous year (X3,t-1) and the tractor demand of the previous year (Yt-1). The models studied for tractor demand in Italy, France and the United States show that the univariate (ARIMA) models have a lower statistical validity than multivariate ones only for France. The multivariate models present a remarkable capacity for explanation even though in the case of Italy and France the tractor price index is not significant while the net farm income is not significant for the United States. Tractor demand, obviously, increases with the increase in the added farm value or income of the farmers in real terms, and lessens with the growth of the real price index of tractors and with the growth of the stock. The models studied and used for the forecasts, however, have limitations because of the high standard errors of the estimate which determines the large confidence interval of the forecasts, especially in the increase of the forecast horizon. The forecast for tractor demand is quantifiable for the short period with acceptable errors (less than 10% according to the standard error of the estimate) but becomes aleatory at medium term (due to the increase in forecast errors with the growth in the forecast horizon), as with all human forecasts.