Boon,

  Can you provide more information, for example, the bipolar model
that you are using: Gummel-Poon, HiCUM, ...  and the test circuit?

The following comments are for the gp bjt.

1)  The dc operating point ft is a rough approximation, gm/2*pi*f*Cbe
     where Cbe is Cpi + Cbe,junction

2) Measurements include the full device model and other factors,
   Rb, Ccb, Re can effect the simulation result.

3) When you measure ft are you measuring at the frequency cross-over
    or extrapolating from the 3dB frequency using 20dB/dec. When
    measuring, the modeler usually extrapolates to avoid undesired
    high frequency effects, perfect de-embedding is hard to achieve.
    Are the characteristics of the Ic/Ib curve "well-behaved" at the
    cross-over frequency?

a) In general, the simulation result should be more correct, if your
    testbench looks like a real circuit.

b) You may want to build a more "realistic" testbench and extract the
   ft from the s-parameters.

c) You may want to verify that the bias conditions in the your test
   circuit match the conditions in your testbench.  

d) You may want to review the circuit performance, for example, is
   the gain lower in the testbench? For example, could Re be acting
   like an emitter degeneration resistor, lowering gain, increasing
    bandwidth?

Finally, remember that even if the ft result is accurate, it may not
reflect you circuits performance. Depending on what you are designing,
fmax may be a better figure of merit.

                                                                    Best Regards,

                                                                       Sheldon