In forensic applications, identification of the source of a printed document can be very critical. However, this task is very challenging, because there are a variety of factors, such as the media type, the age of the printer, and the amount of colorant left in the cartridge that can affect the characteristics of a printed document. For inkjet printers, the availability of different print modes, and the use of variety of ink types adds even more complexity to this process. In this paper, we investigate different texture features of the characters printed by inkjet printers for classification based on text-only documents. We check the in model stability of these features by using a cost function. Finally, we perform stepwise discriminant analysis to reduce the feature set.