Objective To develop and internally validate a prediction model for tinnitus recovery following unilateral cochlear implantation. Design A cross-sectional retrospective study. Setting A questionnaire concerning tinnitus was sent to patients with bilateral severe to profound hearing loss, who underwent unilateral cochlear implantation at the University Medical Center Utrecht, the Netherlands, between 1 January 2006 and 31 December 2015. Participants Of 137 included patients, 87 patients experienced tinnitus preoperatively. Data of these 87 patients were used to develop the prediction model. Primary and secondary outcome measures The outcome of the prediction model was tinnitus recovery. Investigated predictors were: age, gender, duration of deafness, preoperative hearing performance, tinnitus duration, severity and localisation, follow-up duration, localisation of cochlear implant (CI) compared with tinnitus side, surgical approach, insertion depth of the electrode, CI brand and difference in hearing threshold following cochlear implantation. Multivariable backward logistic regression was performed. Missing data were handled using multiple imputation. The performance of the model was assessed by the calibrative and discriminative ability of the model. The prediction model was internally validated using bootstrapping techniques. Results The tinnitus recovery rate was 40%. A lower preoperative Consonant-Vowel-Consonant (CVC) score, unilateral localisation of tinnitus and larger deterioration of residual hearing at 250 Hz revealed to be relevant predictors for tinnitus recovery. The area under the receiver operating characteristics curve (AUC) of the initial model was 0.722 (IQR: 0.703-0.729). After internal validation of this prediction model, the AUC decreased to 0.696 (IQR: 0.667-0.700). Conclusion and relevance Lower preoperative CVC score, unilateral localisation of tinnitus and larger deterioration of residual hearing at 250 Hz were significant predictors for tinnitus recovery following unilateral cochlear implantation. The performance of the model developed in this retrospective study is promising. However, before clinical use of the model, the conduction of a larger prospective study is recommended.