Herbaceous perennials represent a considerable portion of potential biomass feedstocks available for the growing bioenergy industry. Their chemical composition and biomass yields, which are important in determining ethanol potential on an area and mass basis, vary with plant variety and type, environment, and management practices. Therefore, a study was conducted to assess the variability of lignin and carbohydrate content, biomass yields, and theoretical ethanol yields on an area basis among different herbaceous perennial species combinations grown in Minot (2008) and Williston (2008, 2009, and 2010), North Dakota (ND). After wet chemistry compositional analysis was done, the carbohydrate contents were used to determine theoretical ethanol potential on a mass basis. Using the dry-matter yield, the theoretical ethanol yield on an area basis was also calculated for these biomass species. Total carbohydrate content for the biomass samples in Williston and Minot varied from 45 to 61% dry basis. Analysis of Variance (ANOVA) at a= 0.05 showed that carbohydrate content varied between years and environments. Also an interaction plot shows that no biomass species had consistently higher or lower carbohydrate content in the different environments. Switchgrass (Panicum vigatum L.) grown as single species or together with other perennial grasses had higher dry-matter yield and theoretical ethanol yield potential in Williston irrigated plots while mixtures containing intermediate or tall wheatgrass species (Thinopyrum spp.) produced better yields in Minot non-irrigated plots. Variability in theoretical ethanol yield on a mass basis (3.7% coefficient of variation (CV) in Williston and 9.7% CV in Minot) was much less than the variability in dry-matter yields (27.5% CV in Williston and 14.8% CV Minot). Therefore, biomass production is much more important than composition in choosing species to grow for ethanol production. Recently, many studies have focused on developing faster methods to determine biomass composition using near infrared (NIR) spectroscopy. Other NIR models have been developed on single biomass feedstocks but a broad-based model for mixed herbaceous perennials is yet to be developed. Therefore, NIR calibration models for lignin, glucan, and xylan were developed with 65 mixed herbaceous perennial species using a DA 7200 NIR spectrometer (950 - 1,650 nm) and GRAMS statistical software. The models for lignin and xylan had R(2) values of 0.844 and 0.872, respectively, upon validation and are classified as good for quality assurance purposes while glucan model had an R(2) of 0.81 which is considered sufficient for screening. The R(2) and the root mean square error of prediction (RMSEP) results showed that it is possible to develop calibration models to predict chemical composition for mixed perennial biomass when compared with results for models developed for single feedstock by Wolfrum and Sluiter (2009) and Liu et al. (2010). Studying the variability in predicting constituents using NIR spectroscopy over time (hours and days), it was observed that the average CV was between 1.4 to 1.6%. The average CV due to repacking (presentation) alone was 1.3%. The CVs for NIR predictions ranged between 1.4 to 5.7% while for wet chemistry ranged between 3.8 to 13.5%; hence, NIR predictions were more precise than wet chemistry analysis.