Wheat (Triticum aestivum L.) is one of the most important crops cultivated worldwide. Identifying the complete transcriptome of wheat grain could serve as foundation for further study of wheat seed development. However, the relatively large size and the polyploid complexity of the genome have been substantial barriers to molecular genetics and transcriptome analysis of wheat. Alternatively, RNA sequencing has provided some useful information about wheat genes. However, because of the large number of short reads generated by RNA sequencing, factors that are crucial to transcriptome assembly, including software, candidate parameters and assembly strategies, need to be optimized and evaluated for wheat data. In the present study, four cDNA libraries associated with wheat grain development were constructed and sequenced. A total of 14.17 Gb of high-quality reads were obtained and used to assess different assembly strategies. The most successful approach was to filter the reads with Q30 prior to de novo assembly using Trinity, merge the assembled contigs with genes available in wheat cDNA reference data sets, and combine the resulting assembly with an assembly from a reference-based strategy. Using this approach, a relatively accurate and nearly complete transcriptome associated with wheat grain development was obtained, suggesting that this is an effective strategy for generation of a high-quality transcriptome from RNA sequencing data.