Growing attention has been recently given to the use of UAVs for the acquisition of very high spatial resolution multispectral imagery, for agriculture applications. Crop monitoring requires the derivation of state variables such as green leaf area index (GAI) or fraction of Absorbed Photosynthetically Active Radiation (fAPAR) that are later used for decision making. The use of physically based methods for such variables estimation is based on radiative transfer model inversion and therefore requires computing the surface reflectance factor. The classical approach, known as “empirical line” used to calibrate UAV reflectance acquisition relies on the acquisition of images above one or several lambertian reference targets with known reflectance in the considered wavebands before and/or at the end of the flight. A linear relationship is then established between the signal recorded by the sensor and the actual reflectance of the reference target in each waveband. Then, the linear relationship is applied to all the images acquired during the flight to compute the corresponding reflectance factor. However, this method assumes that the illumination conditions are stable during the flight {Smith, 1999 #319}. Illumination conditions address two aspects: (1) the amount of energy that reaches the target, and (2) the directionality of the incoming radiation determined by the sun position and the diffuse fraction which is wavelength dependent. The objective of this study is thus to evaluate the impact of varying illumination conditions (clouds passing during the flight) on the reflectance computation using the empirical line approach. The first experiment addresses the impact of differences in amount of illumination between reference and target acquisition: the camera observes under variable illumination conditions both a vegetation canopy target and reference panels of known reflectance. The resulting reflectance value using the reference panel from a single image calibration is compared to reflectance values derived when the reference panel seen in each image separately is used for calibration. Results show that the reflectance value is very sensitive to the difference between the actual incoming radiation when observing the reference panel and that available when observing the target vegetation. However, this effect can be corrected when normalizing the radiance in each band by the radiance value in a given normalizing band or by the average radiance value over all the bands. Further investigations show that the derivation of canopy state variables is very little dependent on the absolute value of reflectance, but it is rather sensitive to the relative shape of the spectral variation. The effect of the spectral variation of incoming radiation under varying illumination conditions is then addressed by using the normalized reflectance. The ratio of radiances recorded by the camera on the reference is related to the diffuse fraction recorded by a BF5 sensor. Consequences on the estimates of GAI and FAPAR are then discussed. The impact of the diffuse fraction on the canopy BRDF is additionally evaluated by comparison with the concurrent reference panel measurements. Consequences on the canopy state variables are investigated. Finally, conclusions regarding the best measurement strategy including radiometric calibration process, estimation of the diffuse fraction and spectral distribution of incoming radiation are drawn.