Energy-saving has been a central point considered in green architectural designs and researches for decades. However, in the majority of previous studies, the variables in energy-efficient building designs were treated separately by using bivariate-orthogonal methods, and the effects of multivariable interactions were largely neglected. This study focused on the impact and relationships of multiple design factors in three classical glass roof forms (flat, arch and zigzag) in Nanjing, China. Architectural features including glass roof shape flatness, glazing orientation, glazing distribution, glazing material properties were parameterized. The response surface methodology was applied to the simulation design, mathematical model generation and optimization. This study revealed that when compared to the maximum energy demand, up to 87%, 92% and 89% of total energy demand could be saved in the flat, arch and zigzag glass roof designs, respectively. Moreover, in hot summer/cold winter climate regions like the city Nanjing, glazing material properties (U and Tvis) and shape flatness (tan a) had significant impacts on energy demands while glass roof glazing orientation could be neglected. Although generally a glass roof with a flatten shape using low Tvis and U-value material was preferable, multiple optimal solutions with the same desirability of energy-saving were obtained in different settings for arch and zigzag glass roofs. The optimal values for factors U, Tvis and tan a of the zigzag glass roof were distributed over a significant larger range than those of the arch form. The results indicated that the more factors with interactions were taken into considerations, the more possible solutions might be obtained to reach optimal goals, and the restrictions on each factor would be less. The findings regarding the substitutional relations of the research factors have challenged the belief that the relationship between a design factor and energy demand is unidirectional, and therefore may be important for decision-making in energy-saving designs.