The problem of jointly designing the estimation structure and algorithm to infer all or some composition in a six-component distillation column with temperature measurements is addressed. The structure design involves the choices of: (i) modeled and unmodeled compositions, (ii) the number of measurements and their location, and (iii) the innovated-noninnovated state partition. The algorithm is the dynamic data processor that performs the estimation task. The application of the geometric estimation approach (GE), in the light of the column characteristics, yields a tractable procedure to draw the solution of the estimation structure-algorithm design problem, with an estimation scheme that is considerably simpler than previous ones with extended Kalman Filter (EKF). The proposed methodology is applied to a representative six-component case example through simulations, finding that the estimation task can be performed with a three-component reduced model.