The feasibility of near infrared (NIR) reflectance spectroscopy in determining various dry matter constituents such as crude protein, neutral detergent fibre, acid detergent fibre and ash in mixed samples of natural alpine swards has been investigated. Samples were collected in different altitudinal and geographic locations exploited by grazing animals. They were previously analysed by conventional chemical methods, scanned using an NIRSystems 5000 monochromator and spectra were treated using different math pre-treatments and applying different calibration algorithms to achieve the best predictive performances. First and second derivatives of each NIR spectrum were used for all statistical analyses. Step-up, step-wise and modified partial least squares (MPLS) regression methods were applied to develop reliable calibration models between the NIR spectral data and the results of wet analyses. MPLS models fitted chemical data better than linear regression models for CP, ADF and ash, while step-wise worked better for NDF. The results demonstrated that NIR reflectance spectroscopy can be used as a routine testing method to estimate accurately, rapidly and non-destructively three important constituents of mixed alpine forage species, namely crude protein, neutral detergent fibre and ash. For other determinations, such as acid detergent fibre, our results were slightly less accurate and satisfactory but may be useful for separation of samples into groups.