Spectral measurements of thermonuclear (type-I) X-ray bursts from low mass X-ray binaries have been used to measure neutron star (NS) masses and radii. A number of systematic issues affect such measurements and have raised concerns as to the robustness of the methods. We present analysis of the X-ray emission from bursts observed from 4U 1608-52 at various persistent fluxes. We find a strong dependence of the burst properties on the flux and spectral hardness of the persistent emission before burst. Bursts occurring during the low-accretion rate (hard) state exhibit evolution of the black body normalisation consistent with the theoretical predictions of NS atmosphere models. However, bursts occurring during the high-accretion rate (soft) state show roughly constant normalisation, which is inconsistent with the NS atmosphere models and therefore these bursts cannot be easily used to determine NS parameters. We analyse the hard-state burst to put the lower limit on the neutron star radius in 4U 1608-52 of 13 km (for masses 1.2-2.4 M ⊙ ). The best agreement with the theoretical NS mass-radius relations is achieved for source distances in the range 3.1-3.7 kpc. We expect that the radius limit will be 10 per cent lower if spectral models including rapid rotation are used instead.

Neutron star (NS) masses and radii can be estimated from observations of photospheric radiusexpansion X-ray bursts, provided the chemical composition of the photosphere, the spectral colour-correction factors in the observed luminosity range, and the emission area during the bursts are known. By analysing 246 X-ray bursts observed by the Rossi X-ray Timing Explorer from 11 low-mass X-ray binaries, we find a dependence between the persistent spectral properties and the time evolution of the black body normalisation during the bursts. All NS atmosphere models predict that the colour-correction factor decreases in the early cooling phase when the luminosity first drops below the limiting Eddington value, leading to a characteristic pattern of variability in the measured blackbody normalisation. However, the model predictions agree with the observations for most bursts occurring in hard, low-luminosity, island spectral states, but rarely during soft, high-luminosity, banana states. The observed behaviour may be attributed to the accretion flow, which influences cooling of the NS preferentially during the soft state bursts. This result implies that only the bursts occurring in the hard, low-luminosity spectral states can be reliably used for NS mass and radius determination.