The results of a comprehensive experimental and modelling investigation of high-frequency noise (HFN) of Si/SiGe:C and InP/InGaAs heterojunction bipolar transistors (HBTs) demonstrate that the key parameters to correctly model HFN for both HBTs are the self-heating (SH) effect at high collector injection levels, i.e. J C > 2-3 mA μm −2 , and the noise transport time (τ ) which must be included at all bias levels. In contrast, thermal noise contribution of the base-emitter junction of the Si/SiGe:C HBT, associated with the diffusive nature of electron transport at this junction because of the lack of conduction band discontinuity, is only important at operation frequencies lying in the frequency range f > f T /2. The noise analysis indicates that in order to correctly model HFN performances at any injection level and operation frequency, τ must be constant with bias and equal to the addition of base and collector transit times (τ B + τ C ). On the other hand, at high J C levels if SH is neglected, then the amplitude of the equivalent noise resistance (R n ) will be the most impacted noise parameter for both HBTs. Concerning the InP/InGaAs HBT, if SH is off at J C = 5.5 mA μm −2 , the minimum noise figure (NF min ) of the InP HBT will be underestimated by 10%.