Models have been developed to represent the spread of bovine viral diarrhoea virus (BVDV) in cattle herds. Whereas the herd dynamics is well known, biological data are missing to estimate the parameters of the infection process. Our objective was to identify the parameters of the infection process that highly influence the spread of BVDV in a dairy herd. A stochastic compartmental model in discrete time represented BVDV infection in a typical Holstein dairy herd structured into five groups (calves, young versus older heifers, lactating versus dry cows). Model sensitivity was analysed for variations in the probability of birth of persistently infected (P) calves (b(P)), mortality of P animals (m(P)), within- and between-group transmission rates for P and transiently infected (T) animals (respectively, beta(w)(P),beta(b)(P),beta(w)(T),beta(b)(T)). Three to five values were tested per parameter. All possible combinations of parameter values were explored, representing 3840 scenarios with 200 runs for each. Outputs were: virus persistence 1 year after introduction, time needed to reach a probability of 80% for the herd to be virus-free, epidemic size, mean numbers of immune dams carrying a P foetus, of P and of T animals in infected herds. When considered together, m(P) and beta(b)(P) accounted for 40-80% of variance of all outputs; b(P) and beta(w)(T) accounted each for less than 20% of variance; beta(b)(T) and beta(w)(P) accounted for almost no percent of variance of the outputs. Parameters beta(w)(T) and b(P) needed to be more precisely estimated. The influence of m(P) indicated the effectiveness of culling P calves, the influence of beta(b)(P) indicated the role of the herd structure in BVDV spread, whereas the influence of b(P) indicated the possible role of vaccination programs in controlling within-herd BVDV spread.