We have used several metagenomic approaches to assess ecofunctional genes in microbial community associated with biofuel crop rhizospheres with the ultimate goal of determining whether these genes can aid sustainable biomass production, especially on marginal lands. We sampled seven replicates that are composites of several rhizospheres of the three major biofuel crops (corn, switchgrass and miscanthus) and have sequenced each replicate by 1 lane of Illumina sequence (~300 million reads/sample). We used the following analysis methods: (1) Scalable metagenome assembly. We used digital normalization to reduce redundant sequences and partitioning to subdivide the data into smaller units of connected sequences to make assembly of very large data manageable with constant memory. We have assembled a 400 Gb metagenome with 1Tb of memory and got 13.6 million contigs longer than 400bp. (2) Gene-targeted assembly of N-cycle genes. We developed a new tool (Xander) that uses the Hidden Markov Model to guide local assembly. About 1600 unique nifH contigs (nitrogenase genes) were assembled with an average coverage of less than 10, a greater yield than by general assembly approach. (3) Data mining of the shotgun short reads using reference gene models. This method has been applied to16S rRNA gene and compared with 454 amplicon data. The two types of data showed similar community profile at phylum level, but shotgun data also revealed some taxa (Verrucomicrobia) missed by amplicon sequencing primers. Preliminary HMMER search results on the N-cycle gene on data sets showed: (1) large variations between two rhizosphere replicates in both miscanthus and switchgrass. (2) Denitrification genes (narG) are more abundant than nitrogen fixation genes (nifH) and nitrification genes (amoA).