The grazing behaviour and diet selection of a vertebrate herbivore, the domestic sheep (ovis aries L.), feeding in patchy perennial ryegrass (Lolium perenne L.) monocultures was investigated. Heterogeneity or patchiness was created by manipulating (1) the nutritional content of ryegrass, (2) parameters of sward structure e.g. sward height, pseudostem and lamina length, and tiller density, and (3) brightness of ryegrass relative to the background sward. Patches were either fertilised turves transplanted into paddock swards or were created in situ by local fertilisation, trimming etc. The background sward was low in nitrogen (<1.0%N dry weight) compared with average ryegrass leys found on farms in British lowland areas. Sheep preferred to graze in transplanted patches of ryegrass. These patches were structurally similar to the background sward, but were higher in nitrogen content and digestibility as a result of fertilisation. Thus, sheep actively selected more nutritious ryegrass. There was a significant quantitative relationship between brightness and the nitrogen and water soluble carbohydrate content of ryegrass. Darker green ryegrass had a higher nitrogen content and lower water soluble carbohydrate content than lighter green ryegrass. It is suggested that sheep learned to use brightness as a cue in intra-specific forage selection. They had a high preference for short, dark green patches of similar height to the lighter green, background sward. Sheep also preferred to graze in tall patches of ryegrass (>10 cm taller than the background sward) regardless of their nutritional content or brightness relative to the background sward. It is likely that sheep use a multiplicity of physical cues associated with perennial ryegrass in diet selection. The interaction between structural and sward parameters and nutrition content of forage in influencing diet selection requires further investigation. In further experiments, an optimal foraging model, the marginal value theorem was used to make predictions about the behaviour of sheep grazing in monocultures containing highly preferred transplanted patches of tall, fertilised ryegrass. By manipulating the cost of travel between these good ryegrass patches and by detailed measurements of intake within patches by individual sheep, I was able to make quantitative predictions about the time spent grazing in each patch. Travel time and costs were increased by moving good patches further apart and by hobbling sheep to slow down their speed of movement. As predicted by the marginal value theorem, average time spent in a patch was positively correlated with average travel time between patches. For two sheep, the predicted and observed times spent in patches were statistically similar at low travel times, while at higher travel times, observed times were greater than predicted. For a third sheep, all observed times were greater than predicted. Thus overall, there was not a good quantitative fit between the model and the observed behaviour. While incorporation of the difference in the energetic cost of travel relative to that of grazing in a good patch into the model resulted in an increase in predicted optimal patch residence times, the fit between predicted and observed values was not significantly improved. Reasons for this difference between observed and predicted patch residence times are discussed. However, these experiments demonstrated that patch-use foraging models may be more useful than prey models in investigating grazing behaviour of vertebrate herbivores, and that travel time between patches of forage is a previously unidentified constraint of sheep grazing behaviour.