Six documents were analyzed - ABD's [Asian Development Bank] two fisheries policies and four fisheries project documents. Two of the documents were issued in 1980, one in 1991, and three in 1997. The study compared the three early documents with the three 1997 ones. The text were coded using open coding and CDA-[critical discourse analysis] based (linguistic) coding. ADB's discourse practices and its social relations were also examined. Three constructs emerged from the textual analysis, through the combination of open coding and CDA coding - influence, dependence and accountability. In the three 1997 documents, there was more evidence of self-promotional language. Despite references to participation in the 1997 documents, the language used to refer to the people ADB projects are designed to help implies that ADB considers them as dependent. The language used in the 1997 documents were less transparent than that in the earlier documents which resulted in less overt accountability in the discourse analysed. ADB's discourse practices, specifically the preparation and dissemination of documents, changed in 1995 when its Policy on Confidentiality and Disclosure of Information came into effect. The policy stipulated greater participation in preparing ADB discourse, wider dissemination of its discourse, and more transparency and accountability. While there is wider dissemination of documents, the study revealed that the extent to which the other objectives have been obtained is debatable. Analysis of ADB's sociocultural relations showed that it is coming under increasing criticism, and must therefore make more efforts to promote itself. As other parties in its development environment exert pressure on ADB to change, its discourse practices become more open. At the same time, however, its discourse appears to become more opaque. CDA exposed ADB's perception of itself, and of its relationship with its stakeholders; it highlighted the relationship between ADB discourse, its discourse practices and its social relations. Nud.ist is a useful tool for the grounded theory method; however, it might not be the best program for discourse analysis as it cannot search for grammatical categories such as adjectives or passive voice