The three-way decisions model based on incomplete interval-valued information tables is generalized and improved. Firstly, according to the risk preference of decision-makers, a L level similarity relation is defined to measure the similarity degree between different objects in the interval-valued information table. Secondly, based on the two semantics of missing values in incomplete information tables, the interval number is used to obtain the loss function, and then the threshold value required by the three-partition is obtained by Bayesian minimum risk decision theory. Finally, the three-way decision is extended to the mixed interval-valued information tables which is composed of incomplete interval-valued information and loss function matrix, and the key steps and model methods of constructing the three-way decision are studied in detail.