We consider in this work thermal multiphase multicomponent flows in porous media. We derive fully computable a posteriori error estimates for the dual norm of the residual supplemented by a nonconformity evaluation term. The estimators are general and valid for a variety of discretization methods. We also show how to estimate separately the space, time, linearization, and algebraic errors giving the possibility to formulate adaptive stopping and balancing criteria. Moreover, a space–time adaptive mesh refinement algorithm based on the estimators is proposed. We consider the application of the theory to an implicit finite volume scheme with phase-upwind and two-point discretization of diffusive fluxes. Numerical results on an example of real-life thermal oil-recovery in a reservoir simulation illustrate the performance of the refinement strategy and in particular show that a significant gain in term of mesh cells can be achieved.