Flavour compounds vary in physicochemical properties and therefore behave differently in foods with different fat content. The objective was to investigate differences in dynamic flavour intensity in a realistic food system and relate them to two relevant molecular descriptors for flavour compounds. Ice creams with different fat levels (3%, 6% and 12% milk fat) and flavouring (β-ionone (berry), δ-nonalactone (coconut), isopentyl acetate (banana), vanillin (vanilla)) were examined. Approximately iso-intense concentrations (in 12% fat) were selected. Samples were analysed with time-intensity methodology, evaluating perceived melt rate and flavour intensity (trained panel N=12, 3 replicates). Data were analysed by ANOVA, principal component analysis (PCA) and ANOVA partial least squares regression (APLSR). Analyses of data showed faster perceived melt rates and faster increases and decreases in dynamic flavour perception with lower fat levels. Individual flavour compounds were not affected similarly by changes in fat level. Increase and decrease rates of dynamic flavour perception were related to boiling points and hydrophobicity of the flavour compounds. Isopentyl acetate with the lowest boiling point had the fastest increase rate and the slowest decrease rate in intensity. Vanillin with the lowest hydrophobicity index (logP) had the fastest decrease rate. δ-nonalactone and β-ionone with high boiling points and high logP had similar increase rates as vanillin, but had a slower decrease rate in flavour intensity.