Thermodynamic equilibrium computations are usually the most time-consuming component in compositional reservoir flow simulation. A compositional space adaptive tabulation (CSAT) approach was developed as a preconditioner for equation of state (EOS) computations in isothermal compositional simulation. The compositional space is decomposed into sub- and supercritical regions. In the subcritical region, we adaptively parameterize the compositional space using a small number of tie-lines, which are assembled into a table. The critical surface is parameterized and used to identify supercritical compositions. The phase-equilibrium information for a composition is interpolated as a function of pressure using the tie-line table. We extend the CSAT approach to thermal problems. Given an overall composition at a fixed temperature, the boundary between sub- and supercritical pressures is represented by the critical tie-line and the corresponding minimal critical pressure (MCP). A small set of subcritical tie-lines is computed and stored for a given temperature. This process is repeated for the pressure and temperature ranges of interest, and a coarse (regular) tie-line table is constructed. Close to the critical boundary, a refined tie-line table is used. A combination of regular and refined interpolation improves the robustness of the tie-line search procedure and the overall efficiency of the computations. Several challenging problems, including an unstructured heterogeneous discrete fracture field model with 26 components, are used to demonstrate the robustness and efficiency of this general tie-line-based parameterization method. Our results indicate that CSAT provides accurate treatment of the near-critical region. Moreover, the computational efficiency of the method is at least an order of magnitude better than that of standard EOS-based reservoir simulation approaches. We also show the efficiency gains relative to standard techniques as a function of the number of gridblocks in the simulation model.