Looking away from an interlocutor's face during demanding cognitive activity can help adults and children answer challenging mental arithmetic and verbal-reasoning questions (Glenberg, Schroeder, & Robertson, 1998; Phelps, Doherty-Sneddon, & Warnock, 2006). While such “gaze aversion” (GA) is used far less by 5-year-old school children, its use increases dramatically during the first years of primary education, reaching adult levels by 8 years of age (Doherty-Sneddon, Bruce, Bonner, Longbotham, & Doyle, 2002). Furthermore, GA increases with increasing mental demands, with high levels signalling that an individual finds material being discussed challenging but remains engaged with it (Doherty-Sneddon et al., 2002; Doherty-Sneddon & Phelps, 2005). In the current study we investigate whether patterns of gaze and gaze aversion during children's explanations can predict when they are in states of transient knowledge (Karmiloff-Smith 1992; Goldin-Meadow, Kim, & Singer, 1999). In Study 1, thirty-three 6-year-old children completed a balance beam task (Pine & Messer, 2000). Children who improved the representational level of their explanations (Karmiloff-Smith, 1992) of this task with training used more GA than those who did not. Practical implications for teaching and for recognizing transient knowledge states are discussed. In Study 2, fifty-nine 6-year-olds took part and completed a “Time Task” along with periodic teaching intervention to improve their comprehension of telling the time. Some children improved immediately, whereas others did so more gradually. The gradual improvers showed the highest levels of GA, particularly when they were at an intermediate level of performance.