Plant small RNAs are a diverse and complex set of molecules, ranging in length from 21 to 24 nt, involved in a wide range of essential biological processes. High-throughput sequencing is used for the discovery and quantification of small RNAs. However, several biases can occur during the preparation of small RNA libraries, especially using low input RNA. We used two stages of maize anthers to evaluate the performance of seven commercially-available methods for small RNA library construction, using different RNA input amounts. We show that when working with plant material, library construction methods have differing capabilities to capture small RNAs, and that different library construction methods provide better results when applied to the detection of microRNAs, phasiRNAs, or tRNA-derived fragment. We also observed that ligation bias occurs at both ends of miRNAs and phasiRNAs, suggesting that the biased compositions observed in small RNA populations, including nonstoichiometric levels of phasiRNAs within a locus, may reflect a combination of biological and technical influences.