Production Well Testing is one of the major routines in the upstream oil and gas operations. It is an important oil & gas operation as it will be used to track the performance of individual well productivity, production planning, projection for future investment or divestment i.e. drill new wells, shut non-productive wells or execute well intervention. Historically, production well testing is executed using a single-phase meter after the hydrocarbon is separated in a test separator. Alternatively, physical Multiphase Flow Meter (MPFM) could be used either over a test header or at individual well flowline. MPFMs combine several measurement principles such as Gamma ray spectroscopy, capacitance tomography, microwave, and ultrasound to infer phase flow rates. From PETRONAS experience, there are many instances where the MPFM, especially in subsea installation, failed either prematurely or later part of the field life. Virtual Flow Meter (VFM) can be used as a back-up in this case. Traditionally, a virtual flow meter uses a steady state or transient flow simulator that exploit various measurements within the wells and facility such as downhole pressure & temperature gauges, tubing head flowing pressure, differential pressure across the choke, choke position and fluid characteristic to derive the flow rates of oil, water and gas. In this paper, we will compare and evaluate a VFM that uses transient flow simulator against data-driven VFM which uses machine learning technique. Both data-driven and transient flow simulator VFM are evaluated using the same well test data from real producing oil fields. From this work, the team realized there is a potential that both data-driven and transient flow simulator can be combined to provide a complementary benefit as an optimal Virtual Flow Meter and to reduce the overall system uncertainty.