Thermal conductivity (λ) of frozen foods was modelled using meat composition data and temperature (T) dependent λ of solid components. The λ of non-ice/water components showed slight variation with T. Subsequently, the influence of λice = f(T) on the predicted λ of frozen foods was studied using a variety of models with continuous and dispersed phases. The predictive models were tested against experimental data of 13 frozen meat products in T range from -40 to -1°C and compared with the λ data obtained for constant λice (2.21 W/m⋅K). Variable λice = f(T) noticeably improves the performance of models with constant λice, which were under-predicting experimental data, and worsens the models whose predictions originally followed closely to experimental data. The use of λice = f(T) was beneficial to models with the under-predictive performance, particularly below -5°C, while between -1 and -5°C no noticeable changes were spotted.