A novel method for non-destructive intramuscular fat (IMF) estimation via spectral ultrasound backscatter analysis of signals obtained from pig carcasses early post mortem is described. A commercial hand-held ultrasound device (center frequency: 2.7 MHz) was modified to focus the sound beam to the longissimus muscle at the 2nd/3rd last rib. Time-resolved ultrasound backscatter signals of loin muscle were recorded 45 min p.m. on 82 pig carcass sides. Backfat width (d(BF)=18.9±3.8 mm) and muscle attenuation (α(muscle)=.77±.15 dB MHz(-1) cm(-1)) were assessed from the measured pulse echo data. Other propagation properties of skin, backfat and muscle tissue obtained in a previous investigation were incorporated into the signal pre-processing to minimize parameter estimation artifacts. Spectral and cepstral parameters were derived from time-gated backscattered signals measured in the central muscle region. The range of intramuscular fat (IMF) determined by ether extraction was representative for German pig populations (.7%≤IMF(chem)≤3.6%, coefficient of variation CV(IMF(chem))=44.8%). Variations of IMF were associated with variations of backfat width (CV(d(BF))=20.2%), muscle attenuation (CV(α(muscle))=19.3%), and slope of the backscattered amplitude spectrum (CV(m)=28.8%). A full cross validated multiple linear regression model using these parameters resulted in good predictability of IMF(chem) (R(2)=.76, RMSEP=.34%). Among all tested carcasses, 73% could be correctly classified into one of three IMF classes (LOW: <1%, MID: 1-2%, HIGH: >2%). Using a single threshold (2% IMF), about 92% of all carcasses were correctly classified. With respect to the inherent variability of IMF within a single muscle and the different tissue volumes used for the chemical and ultrasound based IMF estimations the remaining prediction errors are acceptable. Compared to previous ultrasound based studies, the number of acoustic parameters used for the IMF prediction could be reduced. Moreover, the used parameters are based on time-of-flight and spectral slope estimations, which are i) more robust with respect to measurement artifacts and ii) have a causal link to structural variations associated with IMF variations in pork loin.