Using previously established knowledge about survey mode preferences distribution in a population can be one of many ways of improving representativeness and quality of data gathered by survey research. Apart from mode preference existence and stability, the main problem concerns the question: which of the sources of information about mode preference could be treated as trustworthy. Because real observed choices are typically treated as better predictors of future choices than declaration only, this paper tries to answer the question: ‘Are the declarations a good predictor of preference in comparison to real choices?’ For this purpose, it uses combined data from 1) the 2015 Polish mixed-mode ESS experiment and 2) data from ESS8 in Poland. The multinomial logistic regression survey preference model includes socio-demographic variables accompanied by declaration/choice control variable. The results suggest significant differences between choices/declarations. Findings could be used to refine the contact strategies used in surveys.