For wood products that contain visible wood it is important to be able to describe the aesthetic properties desired, to measure and communicate them. The aim of this investigation was to shed light on the connections between people's preferences and the physical blend of wood features measured in digital images of Scots pine wood surfaces.  A total of 215 persons from Sweden with different backgrounds were interviewed regarding their preference for ten Scots pine wood surfaces containing knots. Their impressions and preferences were documented by a questionnaire. Many texture features (140 variables) were extracted from grayscale images of the wood surfaces and the connections with the preference data were modeled by partial least square analysis. Results from detail questions concerning what people thought about specific wood features were used as guidelines for the variable extraction. Prediction models for the preference questions were established and reported. For each prediction model, the 30 most robust wood feature variables were sorted out before the modeling. The results show that the wood feature variables varied in importance and all but one model was significant. The most important variables were those that detect different kinds of feature distribution over a wood surface and especially those variables that detect a deviation in center of gravity. The results illuminate the use of subjective preference data regarding the aesthetic properties of wood and give rise to some ideas of how to implement them in a production process.