Abstract OBJECTIVES Annuloplasty rings are routinely used in mitral valve repair (MVr). However, accurate annuloplasty ring size selection is essential to obtain a favourable outcome. Moreover, ring sizing can be challenging in some patients and is highly influenced by surgeons' experience. This study investigated the utility of three-dimensional mitral valve (3D-MV) reconstruction models to predict annuloplasty ring size for MVr. METHODS A total of 150 patients undergoing minimally invasive MVr with annuloplasty ring due to Carpentier type II pathology and who were discharged with none/trace residual mitral regurgitation were included. 3D-MV reconstruction models were created with a semi-automated software package (4D MV Analysis) to quantitate mitral valve geometry. To predict the ring size, univariable and multivariable linear regression analyses were performed. RESULTS Between 3D-MV reconstruction values and implanted ring sizes, the highest correlation coefficients were provided by commissural width (CW) (0.839; P < 0.001), intertrigonal distance (ITD) (0.796; P < 0.001), annulus area (0.782; P < 0.001), anterior mitral leaflet area (0.767; P < 0.001), anterior–posterior diameter (0.679; P < 0.001) and anterior mitral leaflet length (0.515; P < 0.001). In multivariable regression analysis, only CW and ITD were found to be independent predictors of annuloplasty ring size (R2 = 0.743; P < 0.001). The highest level of agreement was achieved with CW and ITD, and 76.6% of patients received a ring with no >1 ring size difference from the predicted ring sizes. CONCLUSIONS 3D-MV reconstruction models can support surgeons in the decision-making process for annuloplasty ring sizing. The present study may be a first step towards accurate annuloplasty ring size prediction using multimodal machine learning decision support.