Different light field cameras (i.e., traditional and focused) can be used for the flame temperature measurement. But it is crucial to investigate which light field camera can provide better reconstruction accuracy for the flame temperature. In this study, numerical simulations were carried out to investigate the reconstruction accuracy of the flame temperature for the different light field cameras. The effects of flame radiation sampling of the light field cameras were described and evaluated. A novel concept of sampling region and sampling angle of the light field camera was proposed to assess the directional accuracy of the sampled rays of each pixel on the photosensor. It has been observed that the traditional light field camera sampled more rays for each pixel, hence the sampled rays of each pixel are approached less accurately from a single direction. The representative sampled ray was defined to obtain the direction of flame radiation. The radiation intensity of each pixel was calculated and indicated that the traditional light field camera sampled less radiation information than the focused light field camera. A non-negative least square (NNLS) algorithm was used to reconstruct the flame temperature. The reconstruction accuracy was also evaluated for the different distances from microlens array (MLA) to the photosensor. The results obtained from the simulations suggested that the focused light field camera performed better in comparison to the traditional light field camera. Experiments were also carried out to reconstruct the temperature distribution of ethylene diffusion flames based on the light field imaging, and to validate the proposed model. (C) 2018 Elsevier Ltd. All rights reserved.