Data on a commercial Duroc pig line were used to estimate the heritability of intramuscular fat content (IMF, n=1313) and its genetic correlations with body weight (BW, n=19862) and ultrasonic backfat thickness (BF, n=19099) at 180 days, weight of hams (WH, n=1305) and number of piglets born alive at first parity (BA, n=5114). IMF was estimated in the gluteus medius muscle of females by near infrared transmittance spectroscopy. Restricted maximum likelihood estimates of variance components were obtained using a multivariate animal model, in which BW and BF were adjusted for age at test and WH and IMF for carcass weight. Estimates of heritability for IMF, BW, BF, WH and BA were 0.57, 0.18, 0.35, 0.41 and 0.06, respectively. Estimates of the genetic correlations of IMF with BW, BF, WH, and BA were 0.34, 0.64, −0.34 and 0.12, respectively. The efficacy of selecting for BW at restrained BF as a strategy for increasing IMF was examined comparing results on IMF in three groups of barrows (n=200) differing in the mid-parent predicted breeding value for BW but not for BF. IMF was determined in the gluteus medius muscle (IMFGM) at 225 days and in the longissimus dorsi muscle (IMFLD) at 160, 180, and 225 days. Responses for BW and BF were in agreement with the selection procedure used to establish the groups (115.4±1.4, 111.1±1.2, and 107.2±1.5 kg, for BW, and 18.9±0.4, 18.9±0.4, and 18.8±0.5 mm, for BF, at 180 days) but IMF did not differ among groups at any muscle and age. The results indicate that, with the estimated correlation structure, selection for BW at restrained BF is not an effective strategy for improving IMF, although favourable expected response scenarios for IMF, BF and BW are possible if IMF is recorded routinely. The predictive ability of the model for IMF was assessed using the regression coefficients of IMFGM and IMFLD on the mid-parent predicted breeding value for IMF. Observed values were more consistent with predicted genetic values for IMFGM (2.17±0.54, P