Résumé : We investigate the nature and the physical properties of z, Y and, J‐dropouts galaxies selected behind the lensing cluster A2667. This field is part of our project aimed at identifying z~7‐10 candidates accessible to spectroscopic studies, based on deep photometry with ESO/VLT HAWK‐I and FORS2 (zYJH and Ks‐band images, AB(3 sigma)~26‐27) on a sample of lensing clusters extracted from our multi‐wavelength combined surveys with SPITZER, HST, and Herschel. We have selected 1 J‐dropout, 8 Y‐drops and 1 z‐drops within the common field of ~33 arcmin2 (effective area once corrected for contamination and lensing dilution). All of them are detected in H and Ks bands in addition to J and/or IRAC 3.6/4.5, with H(AB)~23.4 to 26.5, and have modest magnification factors. Although best‐fit photometric redshifts place all these candidates at high‐z, the contamination by low‐z interlopers is estimated at 50‐75% level based on previous studies, and the comparison with the blank‐field WIRCAM Ultra‐Deep Survey (WUDS). The same result is obtained when photometric redshifts include a luminosity prior, allowing us to remove half of the original sample as likely z~1.7‐3 interlopers with young stellar pulsations and strong extinction. Two additional sources among the remaining sample could be identified at low‐z based on a detection on Herschel data. Another source, our J-drop candidate, was also removed from our original high-z sample using X-Shooter observations.  These low‐z interlopers are not well described by current templates given the large break, and cannot be easily identified based solely on optical and near‐IR photometry. Given the estimated dust extinction and high SFRs, some of them could be also detected in the IR or sub‐mm bands. After correction for likely contaminants, the observed counts at z>7.5 seem to be in agreement with an evolving LF. We are able to constraint the bright end of the LF at z~8 and 9 after correction for contamination and incompleteness. Our results are consistent with previous studies.