Likert-type scaling has been widely used in survey and social science research. This study investigated the effects of choice of scale labels and of semantic compatibility on rating scale responses in Likert-type scaling, via three experiments.Experiment 1 estimated the scale values of four types of scale labels: evaluative, frequency, agreement and competence labels. Scale values obtained from Experiment 1 were used to construct the rating scales for Experiments 2 and 3. Using the modified magnitude estimation method, 60 education students rated evaluative, frequency, agreement and competence labels. Results showed that in general these labels were good descriptors of evaluative, frequency, agreement and competence domains.Experiment 2 examined the effects of centredness and packedness of rating scale labels on rating responses. Using frequency labels, two levels of centredness were crossed with three levels of packedness. 210 education students rated their attitudes towards university student life on a five-point scale questionnaire. Results showed that there was no significant interaction between centredness and packedness, but that there were significant main effects of centredness and packedness.Since there is limited research available on the effects of centredness and of semantic compatibility on rating scale responses, further investigations on these factors should be undertaken. Limitations of the study and directions for future research are also discussed.Experiment 3 examined the effects of semantic compatibility between item content and rating scale labels on rating responses, and the internal consistency of these responses. Using agreement and evaluative labels, two levels of packedness were crossed with two levels of type of scale labels. 160 education students rated their attitudes regarding their academic abilities on a five-point scale questionnaire. Results showed that where semantically compatibility exists between item content and rating scale labels, the mean score was higher for the equal interval scale than that for the right packed scale, but where semantic compatibility does not exist, the mean scores of the equal interval and right packed scales were equal to each other. Reliabilities for the right packed and equal interval scales in the matched condition were also not significantly different from those in the mismatched condition.