We present the UDF05 HST program, which consists of three disjoint fields—NICP12, NICP34, plus the HUDF—with deep ACS (F606W, F775W, and F850LP) and NICMOS (F110W and F160W) imaging. Here we use the ACS data for the NICP12 and HUDF fields to implement a (V − i) − (i − z) selection criterion that allows us to identify a sample of 101 (133) z ~ 5 Lyman break galaxies (LBGs) down to z850 = 28.5 (29.25) mag in NICP12 (HUDF). We construct the rest-frame 1400 Å LBG luminosity function (LF) over the range M1400 = [ − 21.4, − 17.1] , i.e. down to ~0.04L* at z ~ 5, and use Subaru Deep Field results (Yoshida et al. 2006) to constrain our LF at the bright end (M1400 ≥ − 22.2). We show that (1) different assumptions regarding the LBG SED distribution, dust properties, and intergalactic absorption result in a 25% variation in the number density of LBGs at z ~ 5; (2) under consistent assumptions for dust properties and intergalactic absorption, the HUDF is ~30% underdense in z ~ 5 LBGs relative to the NICP12 field, a variation which is well explained by cosmic variance; and (3) the faint-end slope of the LF does not depend on the input parameters, and has a value of α ~ − 1.6, similar to the faint-end slope of the LF of z ~ 3 and z ~ 6 LBGs. Our study therefore supports no variation in the faint end of the LBG LF over the whole redshift range z ~ 3 to z ~ 6. Based on a comparison with semianalytical models, we speculate that the z ~ 5 LBGs might have a top-heavy IMF.