A generic method for predicting the complexation performance of cyclodextrins complexes is desirable. Two quantitative structure–property relationship models are established for the complexation performance of α- and β-cyclodextrins. The datasets consist of 229 α-cyclodextrin and 330 β-cyclodextrin complexes, including conventional organic molecules, ionic organic compounds, and drugs. Internal validation, indicated by the squared correlation coefficient (Q2) of 0.8949 and 0.8954 for α- and β-cyclodextrin models, demonstrates the stability and robustness of these models. For external validation, the squared correlation coefficients of test sets (R2test) are 0.9315 (α-cyclodextrin model) and 0.9171 (β-cyclodextrin model), confirming the predictive ability of both models. Furthermore, the complexation performances of two ionic liquids are measured using UV spectrometric titration for our models’ external prediction. The mean absolute error between measured and calculated values for β-cyclodextrin complexes is 0.20 M−1, indicating the satisfactory transferability of the model. For the α-cyclodextrin model, the mean absolute error of 1.25 M−1 is deemed acceptable. Both the model validations and experimental results underscore the robustness and predictive performance of these models. In conclusion, these developed models could effectively predict binding constants of several types of molecules with cyclodextrins, which can become valuable tools for cyclodextrin design.