The present research sought to identify linguistic markers of sensitive self-disclosure in Twitter for three main purposes: (1) to support the development of software tools that can identify text as sensitive disclosure or not; (2) to contribute to the literature by establishing what is considered more sensitive disclosure in a specific CMC environment, and (3) to contribute to the methodological toolkit for studying sensitive self-disclosure. Two corpora were used in the present research. In Study 1 short messages were collected from Twitter and the site 'Secret Tweet' for comparison. In Study 2 'tweets' were collected and rated on sensitivity by six raters. LIWC and regression analyses were used to identify the linguistic markers of secret tweets (Study 1, 16 markers found) and sensitive self-disclosure (Study 2, 10 markers found). A software tool is developed to illustrate the markers in application. Implications for self-disclosure research, users, design and researchers are discussed.