Abstract: Analytic rating scales are widely used for assessing interpreting.However, weighting schemes for assessing interpreting reported in previous studies have been largely conceptual and generally pre-determined.Research that investigates weighting based on empirical interpreting assessment data remains scant.And few studies to date have attempted to differentiate between language directions when it comes to weighting.To fill this gap, this study adopts a data-driven approach to exploring weighting schemes for assessing Chinese to English bi-directional interpreting performance.A total of eight raters were invited to evaluate 50 Chinese to English (C-E) interpretations and 50 English to Chinese (E-C) interpretations by trainee interpreters, using an analytic rating scale and a holistic rating scale.Data analysis suggested that: (1) fidelity was the predominant criterion in predicting the candidate's interpreting performance, regardless of interpreting direction (β₁=.351 for C-E interpreting; β₁=.593 for E-C interpreting); (2) delivery came second among the three assessment criteria, regardless of interpreting direction (β₃ =.345 for C-E interpreting; β₃ =.381 for E-C interpreting), and (3) language contributed up to 32.5 percent (β₂=.325) of the variance in the candidate's interpreting performance in the C-E direction, whereas its predictive power on interpreting performance failed to be detected in the E-C interpreting direction due to statistical concerns.Implications of the findings for interpreter training and for the development and validation of assessment tools for interpreting performance are discussed at the end.