There is a growing demand for genome-scale metabolic reconstructions for plants, fueled by the need to understand the metabolic basis of crop yield and by progress in genome and transcriptome sequencing. Methods are also required to enable the interpretation of plant transcriptome data to study how cellular metabolic activity varies under different growth conditions or even within different organs, tissues, and developmental stages. Such methods depend extensively on the accuracy with which genes have been mapped to the biochemical reactions in the plant metabolic pathways. Errors in these mappings lead to metabolic reconstructions with an inflated number of reactions and possible generation of unreliable metabolic phenotype predictions. Here we introduce a new evidence-based genome-scale metabolic reconstruction of maize, with significant improvements in the quality of the gene-reaction associations included within our model. We also present a new approach for applying our model to predict active metabolic genes based on transcriptome data. This method includes a minimal set of reactions associated with low expression genes to enable activity of a maximum number of reactions associated with high expression genes. We apply this method to construct an organ-specific model for the maize leaf, and tissue specific models for maize embryo and endosperm cells. We validate our models using fluxomics data for the endosperm and embryo, demonstrating an improved capacity of our models to fit the available fluxomics data. All models are publicly available via the DOE Systems Biology Knowledgebase and PlantSEED, and our new method is generally applicable for analysis transcript profiles from any plant, paving the way for further in silico studies with a wide variety of plant genomes.