The effective temperature scale of FGK stars, especially at the lowest metallicities remains a major problem in the chemical abundance analysis of metal-poor stars. We present a new implementation of the infrared flux method (IRFM) using the 2MASS catalogue. We computed the theoretical quantities in the 2MASS JHK
s filters by integrating theoretical fluxes computed from ATLAS models, and compare them directly with the observed 2MASS JHK
s magnitudes. This is the main difference between our implementation of the IRFM and that of Ramirez & Melendez (
2005ApJ...626..446R, hereafter RM05), since to introduce new stars at the lowest metallicities they transform the 2MASS JHK
s magnitudes into the TCS photometric system. We merge in our sample the stars from Alonso et al. (
1996A&AS..117..227A, hereafter AAM96;
1999A&AS..139..335A, hereafter AAM99), and other studies to appropriately cover a wide range of metallicities, ending up with 555 dwarf and subgiant field stars and 264 giant field stars. We derived a new bolometric flux calibration using the available Johnson-Cousins UBV(RI)
C and the 2MASS JHK
s photometry. We also computed new T
eff versus colour empirical calibrations using our extended sample of stars. We derived effectives temperatures for almost all the stars in the AAM96 and AAM99 samples and find that our scales of temperature are hotter by ∼64K (σ
T=104K, N=332 dwarfs) and ∼54K with a σ
T=131K (N=202 giants). The same comparison with the sample of RM05 for stars with [Fe/H] ←2.5 provides a difference of ~-87K (σ
T=194K, N=12 dwarf stars) and ∼61K (σ
T=62K, N=18 giant stars). Our temperature scale is slightly hotter than that of AAM96 and RM05 for metal-rich dwarf stars but cooler than that of RM05 for metal-poor dwarfs. We have performed an fully self-consistent IRFM in the 2MASS photometric system. For those who wish to use 2MASS photometry and colour-temperature calibrations to derive effective temperatures, especially for metal-poor stars, we recommend our calibrations over others available in the literature. In our implementation we avoid the transformation of the 2MASS JHK
s magnitudes to a different photometric system and thus fully exploit the excellent internal consistency of the 2MASS photometric system.
