To assess dry matter intake (DMI) in grazing ruminants, techniques have been developed with chromium or ytterbium oxide as inert markers. These markers allow to estimate total faecal output quite accurately, but their analysis is costly and laborious so number of assays is generally limited. Some attempts to use Polyethylene Glycol (PEG, MW6000) as a faecal marker have already been mentioned in literature. NIRS estimation of PEG has been proposed because the reference method (turbidimetry) is tedious. A limitation for the use of NIRS is that faeces characteristics depend closely on the nature of the feed eaten by animals. Therefore calibrations based on a unique nature of faeces are not robust and fail to predict faeces PEG concentrations in animals fed with other diets. The aim of the present experiment is to develop models that predict PEG concentration in faecal output of sheep fed with various diets. A total of 585 samples was obtained by supplementation of 11 faeces with PEG added as a solution in the range 0 - 200 g.kg-1 DM (by 5 or 10 g.kg-1 step). Faeces samples originated from sheep flocks of several breeds and fed with different diets (total mixed diet, dehydrated diet, rangelands, improved pasture). Samples were ground (1 mm sieve) and diffuse reflectance spectra were taken in ring cups on a FOSS NIRSystem 6500. Calibration was carried out, using the modified partial least-squares method (MPLS, WINISI), with wavelengths in the 1100-2500 nm range after mathematical pre-processing (SNV and detrend, 2nd derivative). Equations developed with the whole concentration range (0-200 g.kg-1 DM) on 545 samples (prepared from 10 different faeces) had R2=0.99 and SECV=4.3. However it was noticed that equations developed with narrower concentration ranges were more accurate. The SECV obtained with range scales from 0 to 60, 80, 100, 120 or 140 g PEG g.kg-1 were 1.7, 2.3, 2.4, 2.8 and 3.2 g.kg-1 respectively. The 11th series of samples (not used for calibration) was use