A multi-scale reliability optimization method for honeycomb sandwich composites is proposed based on interval theory. We first develop parameterized finite element models at the microscale for both the face sheets and the core, using a generalized representative volume cell approach to capture the multi-scale characteristics of the material. These models yield the effective mechanical properties of the constituents. We then integrate these homogenized properties into a macroscopic parameterized finite element model of the sandwich structure under three-point bending to evaluate its global mechanical response. Next, we construct a multi-scale reliability optimization model that incorporates interval uncertainties and uses the possibility level of the bending performance as a constraint. To improve computational efficiency, we perform global sensitivity analysis based on interval theory to identify key design and uncertainty parameters. This enables us to reduce the problem"s dimensionality and formulate a more efficient reliability optimization model. Finally, we verify the proposed method through numerical simulations. The results show that, under a 60% possibility level, the optimized design reduces structural weight by approximately 16% and lowers cost by about 22%, while maintaining comparable bending and shear stiffness.