The aim of this paper was to assess which components of landscape structure in lowland English farmland could best predict patterns m bird community composition, total bird density and densities of 12 bird species As far as possible, predictors were sought that were consistent m their effects over a 20 yr period The data were derived from mapping censuses in 1966, 1970, 1974, 1978, 1982 and 1986 with yearly sample sizes ranging from 41 to 51 plots, of minimum size 50 ha Landscape composition was quantified in terms of 10 landscape variables representing broad structural (e g density of woodland) and crop attributes (e g proportion of area under cereals) plus altitude Gradients in bird community composition were defined as plot scores on axis 1 of a detrended correspondence analysis (AXIS1) This gradient consistently represented a gradient from plots rich in woodland bird species to ones rich in field-feeding and wetland species Multiple regression was used to build models containing 2-, 3- and 4-vanables Two approaches were used to combine data from the six years and models based on the single best subset in each year were found to be most consistent with respect to predictor selection Predictors were tested by building models based on each combination of five years and assessing how well the predictors explained variation in the bird variables for the sixth, independent year The predictors were relatively successful in explaining variations in densities of wren, robin, willow warbler, blue tit, great tit and chaffinch but especially AXIS1 which was explained in terms of woodland density, field size and altitude The predictors were unsuccessful in explaining total bird density, song thrush density and whitethroat density Structural attributes of landscape (especially density of hedgerow and woodland, but also rivers and ponds) were frequent predictors of species densities, but variables relating to farming system were not, with the exception of skylark and yellowhammer which were positively associated with extent of cereal crops Implications for landscape design and management, and the relative contribution of woodland and hedgerows to bird communities m farmland landscapes are discussed There was considerable variation between individual years which suggests that models of bird-habitat relationships based on single years may have limited general application