Diagnosis of tuberculosis (TB) in children remains challenging due to unspecific clinical presentation and low bacillary load. In low TB incidence countries, most cases are diagnosed by a contact screening strategy after exposure to an index TB case. Due to the severity of TB in young children, the priority is to determine whether a child is infected or not, whereas differential diagnosis between active TB (aTB) and latent TB constitutes a second step. In Belgium, a low TB incidence country, we prospectively included 47 children with a defined M. tuberculosis infection status (12 children with aTB, 18 with latent TB, and 17 uninfected) (exploratory cohort), and determined the optimal combinations of cytokines secreted by their peripheral blood mononuclear cells in response to a 5-days in vitro stimulation with four different mycobacterial antigens, in an attempt to classify the children according to their infectious status. Correct identification of all infected children was obtained by several combinations of two purified protein derivative (PPD)-induced cytokines (IFN-γ and either GM-CSF, MIP-1α, sCD40L or TNF-α), or by combining PPD-induced IFN-γ with culture-filtrate protein-10 (CFP-10)-induced TNF-α. Alternatively, combining CFP-10-induced TNF-α and IP-10 with heparin-binding haemagglutinin (HBHA)-induced-IFN-γ was more effective in testing recently BCG-vaccinated children or those suspected to be infected with non-tuberculous mycobacteria, providing a correct classification of 97% of the M. tuberculosis-infected children. This combination also correctly classified 98% of the children from a validation cohort comprising 40 M. tuberculosis infected children and 20 non-infected children. Further differentiation between aTB and children with latent TB was more difficult. Combining ESAT-6-induced MIP1-α and IP-10, CFP-10-induced MIG, and HBHA-induced MIG provided a correct classification of 77% of the children from the exploratory cohort but only of 57.5% of those from the validation cohort. We conclude that combining the measurement of 2–4 cytokines induced by three different mycobacterial antigens allows an excellent identification of M. tuberculosis-infected children, whereas differentiating children with aTB from those with latent TB remains far from perfect.