Mindfulness has been applied as an effective therapy in clinical settings to treat psychological symptoms (e.g., depression, stress, anxiety) and also in many other contexts (e.g., the workplace, educational contexts, and sports). Accurate evaluation of mindfulness-based training or mindfulness research requires distinguishing between state and trait changes. The failure to distinguish trait from state in a mindfulness measurement may confuse the assessment results of neurophysiological and psychological mindfulness studies, as well as mindfulness-based treatments, because mindfulness can be understood as either a state or a trait. The Five Facet Mindfulness Questionnaire (FFMQ) is the most widely used multidimentional assessment tool of dispositional mindfulness, comprised of five subscales: “act with awareness”, “describe”, “nonjudge”, “nonreact”, and “observe”. Its short version, consisting of 18 items (FFMQ-18), was proposed by examining and comparing the existing short versions of the FFMQ. However, the ability of the FFMQ to accurately capture stable aspects of mindfulness has not been studied thoroughly, using robust methodology. The generalisability theory (G-theory) is the most suitable and robust statistical approach to differentiate and measure these aspects. In this study, G-theory was applied to distinguish between dynamic (state) and enduring (trait) aspects of mindfulness and to evaluate the reliability of the FFMQ and the FFMQ-18 over time. The study used data from 83 participants, who completed the test on three occasions separated by 2-week intervals. The full 39-item FFMQ and the shorter FFMQ-18 version proved to be very reliable in measuring trait mindfulness, with G coefficients of 0.89 and 0.75, respectively, while individual facet subscales of the FFMQ appeared to be less reliable in measuring trait mindfulness. Subsequent analyses attempted to combine the FFMQ items that were least stable over time to form a measure of state mindfulness. However, these did not result in acceptable psychometric properties for such a state subscale. The findings of this study indicate that a reliable assessment of stable aspects of mindfulness can be achieved by using the full FFMQ scale or its short FFMQ-18 version, with scores that can be generalised across the sample population and assessment occasions. The scores obtained on individual facet subscales of the FFMQ predominantly measure trait mindfulness but their reliability is affected by measurement error, resulting from interaction between persons, items and occasions.