The effect of advanced resonance self-shielding models incorporated in the developmental version of the DRAGON code on estimation of reactivity coefficients of a typical CANDU-6 lattice is evaluated. The advanced self-shielding models are based on either equivalence in the dilution model or on a subgroup approach. Under equivalence in dilution models, the generalized Stamm'ler model was used with or without Riemann integration and Nordheim model. Among the subgroup approaches, the Ribon extended and the statistical self-shielding models were used. The Ribon extended self-shielding model uses mathematical probability tables, while the statistical self-shielding model uses physical probability tables. The analysis focused on four important transients, which include the fuel temperature coefficient, coolant void reactivity, pressure tube ingression, and calandria tube ingression. Four burnup stages for estimation of reactivity have been identified. To benchmark the results obtained using DRAGON, the results obtained were compared with those of MCNP5. These analyses indicated that, of all the self-shielding models, the resonance self-shielding model based on the subgroup approach using physical probability tables seems to perform well for all situations and can be recommended for CANDU-6 analyses using the code DRAGON.