Journal of Agricultural Science, Cambridge (2001), 136, 111–117. Printed in the United Kingdom  2001 Cambridge University Press 111 Statistical modelling of grazing preference of sheep when presented with a range of plant types D.REAL  ,  *, I. L. GORDON  ,   J.HODGSON  ,   Department of Plant Science, Massey University, PO Box 11.222, Palmerston North, New Zealand  INIA Tacuarembo  , Ruta 5 km 386, CP 45000, Tacuarembo  , Uruguay  Institute of Molecular Biosciences, Massey University, Palmerston North, New Zealand  Institute of Natural Resources, Massey University, Palmerston North, New Zealand (Revised MS received 22 June 2000) SUMMARY A statistical model of the grazing preference of sheep is presented for the evaluation of spaced plants in small plots for plant breeding purposes. Plants are located randomly to reduce the effect of external factors on diet selection, and to differentiate between discriminatory and random grazing. Consistency of discrimination among grazings and sites (Massey University, New Zealand and INIA La Estanzuela, Uruguay) was tested. The statistical design and subsequent analyses considered all sources of variation to minimize error, and to separate genetic effects from environment effects. Clonal replicates were used to enhance error-control, and hence the precision of heritability estimates, as most characters are inherited quantitatively. Post-grazing leafiness is considered the best character to select and breed for animal preference in a red clover germplasm. INTRODUCTION The history of breeding pasture plants is relatively short, but in those forage species in an advanced stage of domestication, objectives other than increasing yield are becoming more relevant. These include persistence, and production under limiting resources (e.g. high competition or poor growing environment) and high grazing pressure. To evaluate a large number of lines or selections, as is the case in early stages of any forage breeding programme, cutting is the easiest and most economical method of defoliation. However, many researchers have expressed concern about the validity of the results because they omit realistic animal effects on the pasture (Hodgson 1981 ; Evans et al. 1992; Swift et al. 1992 ; van Santen 1992). The grazing animal can alter the development of forage species directly by defoliation, treading and excreta return, and indirectly by changing the sward structure and micro-environment (Curll & Wilkins 1983 ; Grant & Marriott 1994). All of these are arguments against having the grazing animal in a breeding nursery, because they all increase errors and reduce heritability estimates. However, despite all the problems, two aspects of plantanimal interaction are important in selecting plants for use in pasture grazing * To whom all correspondence should be addressed. Email : drealinia.org.uy systems : (1) preferential defoliation and level of intake and (2) tolerance of defoliation, and recovery growth. Preference is a complex phenomenon determined by the animal, the plants and the environment in which the plant or plant-part discrimination occurs (Marten 1978). For this reason, it is important to identify firstly a suitable method for determining preference, and secondly plant characteristics with high genetic correlation to preference. The chosen characters not only need to reflect animal reaction, but also to have as high a heritability as possible. This paper outlines a statistical model of grazing preference in forage breeding nurseries. The efficacy of the proposal is evaluated with quantitative genetics from a red clover (Trifolium pratense L.) experiment. MATERIALS AND METHODS The field and statistical design is a randomized complete block in treatments (populations), with subsampling (plants within populations, and ramets within plants). Physically, the plants of the exper- imental units (plots) are randomly scattered rather than contiguous. We propose to call this the ‘ diffuse- plot RCB ’ for cafeteria sampling by animals. The model (see below) allows for repeated grazing events, treating grazing times as a split-plot-in-time after checking for correlated time effects (Gill 1986).