The chapter proposes a concept for controlling the quality of recognition of administrative documents. Features of administrative documents are a limited number of words of static text and many similar documents in streams of digitized pages. Estimates of recognition quality are provided based on statistics of recognition errors, as well as controls of the quality of detection of words and text blocks. An example of a noisy image is given, for which different estimates of recognition quality differ significantly. A description of experiments conducted on our own dataset is provided, in which differences in recognition quality controls are also observed. The proposed concept is based on mapping a set of document recognition results onto a set of data processing system operations. The concept provides an economic control of the quality of recognition, based on the cost of a missed error and the cost of operator work. The proposed concept does not negate the need to use other methods for controlling the quality of recognition, for example, at the stage of training and validation of recognition. The proposed concept allows for effective planning of implementation, and support of document recognition systems. The proposed concept can be applied in modern CAD systems to analyze the content of recognized textual files.