Experimental data on RC members subjected to pure torsion is compiled from published studies so far. Interestingly, the past studies have concentrated largely on members with small to moderate concrete covers, with handful data on members with thick concrete cover. Widely accepted torsion models including the diagonal compression field theory (CFT), softened truss model (STM) and softened membrane model for torsion (SMMT) are verified on the domains of experimental data missing-in specimens with thick concrete cover. With increasing demand for durability design of RC structures use of thick concrete cover, say up to 75 mm, is being introduced in practice. Recent experimental investigation by the authors demonstrated that spalling of concrete cover, particularly in cases of thick covers, significantly impacts the torsional behavior of RC members. The present study uses these set of experiments to look at the response prediction capability of the existing advanced models and assess the reliability of existing concrete cover spalling theories. For cases with thick cover, the existing models either did not adequately capture the ultimate capacity or erroneously predicted the torsional behavior of the members due to the different mechanics between RC beams with thin and thick covers. Similarly, the spalling theories failed to fully explain the physically observed spalling behavior. Guided by the recent experimental observations and the apparent gaps, the study provides a rational theory for the spalling of concrete cover. The initiation and gradual evolution of concrete cover spalling is traced by observing the acting spalling moment and spalling resistance formulated in the proposed spalling model. The proposed spalling model inherently assumes members susceptibility to spalling of cover concrete and is governed by, the thickness of the concrete cover, tensile strength of concrete, presence of rebar cage and their location, and size of the member. The theory is unified and can explain the spalling of cover due to torsion as well as shear. Its capability is examined by integrating the approach into existing truss model. When compared with state-of-the-art models, the proposed method provides consistent prediction with a relatively smaller scatter, for cases with thick concrete cover as well.