Perkins [1] argues that smoking cue-induced craving has not been demonstrated to have clinical predictive utility, and concludes that it is premature to focus upon this model until it has been demonstrated to predict smoking persistence or other indices of dependence. However, whether this apparent failure of the model is due to limitations of the model itself, or other features of laboratory studies of smoking behavior, is only partially addressed. The concept of ‘craving’, and the use of smokers who are not currently seeking treatment as participants, may be equally problematic. Failure to address these issues may invalidate much laboratory research and limit the applicability of any findings to clinical studies. Despite being used as the primary outcome measure in most laboratory studies, the meaning of subjective craving remains controversial and poorly understood. While it is widely assumed that cravings may precipitate relapse to smoking, measures of self-reported craving predict relapse only weakly, particularly when measured at a single time-point, but even when assessed at regular intervals in ambulatory monitoring studies [2]. The global construct of subjective craving may therefore be of limited utility in predicting relapse; by extension, laboratory studies which focus upon craving may not translate to the real world. An important distinction may be between what might be termed tonic craving (i.e. background levels) and phasic craving (i.e. episodic spikes) [3], the former stimulated primarily by withdrawal states and the latter primarily by cues associated with tobacco use. Intense, brief episodes of craving can occur even after background levels have diminished once withdrawal symptomatology resolves [4], and variability in craving may be more predictive than mean level [5]. The true situation is likely to be more complicated than a simple distinction between tonic and phasic craving, with interactions between craving classes (e.g. conditioned emotional states), but this framework provides a starting point that allows the measurement of multiple components of craving, and at different levels of temporal resolution. It remains to be seen whether parsing these components of craving will enhance the validity of laboratory studies. An alternative is to use implicit biobehavioural measures: smokers show increased vigilance for smoking-related cues (sometimes referred to as attentional bias) compared with non-smokers [6], and this may predict relapse following cessation [7], while measures of smoking topography have also been reported to predict relapse in adolescents [8] and adults [9]. However, these measures often correlate only weakly with self-reported craving [10], and their value in predicting clinically relevant outcomes such as relapse to smoking has not yet been established fully. Another difficulty may be the use of short-term abstinence as a model for smoking cessation. Studies of this kind ‘. . . usually involve smokers who are not interested in quitting permanently but participate for monetary payment’ ([1], p. 2). Biologically, the use of acute abstinence to model withdrawal seems plausible, given the short half-life of nicotine. However, there are important differences between this withdrawal state and that associated with a cessation attempt. In the former case the smoker knows that it will be possible to smoke again shortly, and perceived availability has been shown to modify cue-reactivity. Craving increases more when there is an expectation of an opportunity to smoke [11], although this may depend upon degree of dependence [12], while neural response to cigarette cues varies according to perceived availability [13]. Cue-induced craving has been found to be associated with various proxy indices of tobacco dependence, but not severity of dependence (a key index of relapse risk) [1]. However, laboratory measures unrelated to cue-reactivity do seem to predict relapse. These include measures of physical distress tolerance (such as breath-holding) that have been found to offer predictive utility beyond known risk factors for early smoking lapses [14]. More research is needed to evaluate the possibility that ‘various non-self report correlates of craving in response to cues are stronger predictors of relapse than self-reported craving’ ([15], p. 4). Promising targets include cognitive processing measures such as those mentioned earlier, neural imaging of reactivity to cues, and especially acute changes in smoking behavior (i.e. aspects of smoking topography, such as depth of inhalation) as promising directions to assess cue-reactivity. MM has received nicotine replacement products from GlaxoSmithKline and Pfizer Inc. for distribution to study participants. BH has provided consultancy to Pinney Associates, subcontracted by GlaxoSmithKline and Pfizer Inc.