Measurements of the diffusion length L for triplet excitons in small molecular-weight organic semiconductors are commonly carried out using a technique in which a phosphorescent-doped probe layer is set in the vicinity of a supposed exciton generation zone. However, analyses commonly used to retrieve L ignore microcavity effects that may induce a strong modulation of the emitted light as the position of the exciton probe is shifted. The present paper investigates in detail how this technique may be improved to obtain more accurate results for L. The example of 4,4'-bis(carbazol-9-yl)1,1'-biphenyl (CBP) is taken, for which a triplet diffusion length of L=16 ± 4 nm (at 3 mA/cm 2 ) is inferred from experiments. The influence of triplet-triplet annihilation, responsible for an apparent decrease of L at high current densities, is theoretically investigated, as well as the 'invasiveness' of the thin probe layer on the exciton distribution. The interplay of microcavity effects and direct recombinations is demonstrated experimentally with the archetypal trilayer structure [N,N'-bis(naphthalen-1-yl)-N,N'-bis(phenyl)]-4,4'-diaminobiphenyl (NPB)/CBP/ 2,9-dimethyl-4,7-diphenyl-1,10-phenanthroline (named bathocuproine, BCP). It is shown that in this device holes do cross the NPB/CBP junction, without the assistance of electrons and despite the high energetic barrier imposed by the shift between the HOMO levels. The use of the variablethickness doped layer technique in this case is then discussed. Finally, some guidelines are given for improving the measure of the diffusion length of triplet excitons in operational OLEDs, applicable to virtually any small molecular-weight material.