Context :

Aims : We present the first large , unbiased sample of Lyman Break Galaxies ( LBGs ) at z \sim 1 .
Far ultraviolet-dropout ( 1530 Ã ) galaxies in the Chandra Deep Field South have been selected using GALEX data .
This first large sample in the z \sim 1 universe provides us with a high quality reference sample of LBGs .

Methods : We analyzed the sample from the UV to the IR using GALEX , SPITZER , ESO and HST data .

Results : The morphology ( obtained from GOODS data ) of 75 % of our LBGs is consistent with a disk .
The vast majority of LBGs with an IR detection are also Luminous Infrared Galaxies ( LIRGs ) .
As a class , the galaxies not detected at 24 \mu m are an order of magnitude fainter relative to the UV compared with those detected individually , suggesting that there may be two types of behavior within the sample .
For the IR-bright galaxies , there is an apparent upper limit for the UV dust attenuation and this upper limit is anti-correlated with the observed UV luminosity .
Previous estimates of dust attenuations based on the ultraviolet slope are compared to new ones based on the FIR/UV ratio ( for LBGs detected at 24 \mu m ) , which is usually a more reliable estimator .
Depending on the calibration we use to estimate the total IR luminosity , \beta -based attenuations A _ { FUV } are larger by 0.2 to 0.6 mag .
than the ones estimated from FIR/UV ratio .
Finally , for IR-bright LBGs , median estimated \beta -based SFRs are 2-3 times larger than the total SFRs estimated as SFR _ { TOT } = SFR _ { UV } + SFR _ { IR } while IR-based SFRs provide values below SFR _ { TOT } by 15 - 20 % .
We use a stacking method to statistically constrain the 24 \mu m flux of LBGs non individually detected .
The results suggest that these LBGs do not contain large amounts of dust .

Conclusions :