Assessment of body composition is fundamental in diagnosis and treatment of anorexia nervosa (AN). The gold standard dual-energy X-ray absorptiometry (DXA) is expensive and not universally available. Bioelectrical impedance analysis (BIA) is a non-invasive, inexpensive method relative to DXA. We compared DXA and BIA in the assessment of fat-free mass (FFM), fat mass (FM), and body fat percentage (BF%) in women with AN upon admission (ANT1) and discharge (ANT2) from an inpatient specialist unit with a referent healthy control (HC) group. The study population consisted of 31 ANT1, 25 ANT2, and 52 HC women with median age of 21 years. Body composition was measured by DXA and Tanita foot-to-foot BIA. Comparison between the two methods was done using Bland-Altman analysis, Pearson's correlation coefficient, Lin's concordance correlation coefficient, and linear regression. The mean difference (bias) in FM and BF% values obtained by DXA and BIA in ANT1 (FM: +1.01 kg, BF%: +2.26%) and ANT2 (FM: +1.49 kg, BF%: +1.66%) were comparable to HC (FM: -1.32 kg, BF%: -2.29%) although in opposite directions. Less bias was observed in FFM values in ANT1 (-0.46 kg) and ANT2 (-0.86 kg) than in HC (+2.03 kg); however, the limits of agreement between the two methods were wider in ANT1 and ANT2 than in HC for all measurements. No association was observed between age, percentage of total body water, and the time spent on the inpatient specialist unit with the difference in estimates of body composition between DXA and BIA. Comparison of DXA and BIA suggests that DXA should remain the gold standard for measuring body composition; the development of more specific BIA equations is required to improve validity and precision of BIA in patients with AN. Despite ease and cost in both BIA access and operation, the suitability of BIA in a low bodyweight eating disorders population remains questionable.