Background <p>Adolescence is a critical period for the initiation of alcohol use. It is uncertain how predictors of alcohol onset, such as Theory of Planned Behavior (TPB) constructs, change in response to alcohol onset. This study tests for longitudinal changes in TPB constructs during adolescents’ onset of alcohol consumption.</p> Method <p>Data were collected longitudinally in three waves (T1, T2, T3) in Norway. Participants (<i>N</i> = 1,548; <i>M</i>_age = 13.54, 51% girls, 49% boys) reported alcohol use and TPB constructs (attitudes; subjective norms (SN); Perceived Behavioural Control (PBC); intentions) at each wave. Multilevel modelling was used to test whether the constructs changed over time, and intraclass correlation coefficients were used to assess the total variance explained by differences between individuals and waves. In addition, we tested different lagged versions of the TPB model, to determine whether alcohol use at wave 3 was linked to TPB constructs at earlier waves.</p> Results <p>Alcohol consumption and the TPB constructs varied between waves (58% − 72%). Alcohol use, attitudes and friends-related SN significantly increased from T1 to T2 and from T1 to T3. Intentions and parent-related SN significantly increased only from T1 to T3. PBC significantly decreased only from T1 to T3. Attitudes changed the most, and PBC changed the least over time. Tests of lagged models showed that alcohol use was predicted by constructs measured at earlier waves.</p> Conclusion <p>Adolescent alcohol use was mirrored by changes in attitudes and friends-related SN, whereas changes in intentions, parents-related SN and PBC took longer to manifest. In addition, constructs assessed early in adolescence retained their ability to predict alcohol use later in adolescence, although intentions measured closer to the actual behaviour showed higher accuracy in predicting alcohol use. These findings can be used to determine the optimal timing to intervene in different constructs to delay adolescent drinking onset.</p>