Dried blood spot (DBS) metabolomics has numerous applications in newborn health screening, exposomics, and biomonitoring of environmental chemicals in pregnant women and the elderly. However, accurate metabolite quantification is hindered by several challenges: notably the “hematocrit effect” and unknown blood-spotting volumes. Different techniques have been employed to overcome these issues but there is no consensus on the optimal normalization method for DBS metabolomics, and in some cases no normalization is used. We compared five normalization methods (hemoglobin (Hb), specific gravity (SG), protein, spot weight, potassium (K+)) to unnormalized data, and assessed sex-related differences in the DBS metabolome in 21 adults (group 1, n = 10 males, n = 11 females). The performance of each normalization method was evaluated using multiple criteria: (a) reduction of intragroup variation (pooled median absolute deviation, pooled estimate of variance, pooled coefficient of variation, NMDS and principal component analysis), (b) effect on differential metabolic analysis (dendrogram, heatmap, p-value distribution), and (c) influence on classification accuracy (partial least squares discriminant analysis, sparse partial least squares discriminant analysis error rates, receiver operating curve, random forest out of bag error rate). Our results revealed that Hb normalization outperformed all the other methods based on the three criteria and 13 different parameters; the performance of Hb was further demonstrated in an independent group of DBS from 18 neonates (group 2, n = 9 males, n = 9 females). Furthermore, we showed that SG and Hb are correlated in adults (rs = 0.86, p < 0.001), and validated this relationship in an independent group of 18 neonates and infants (group 3) (rs = 0.84, p < 0.001). Using the equation, SG = −0.4814Hb2 + 2.44Hb + 0.005, SG can be used as a surrogate for normalization by Hb. This is the first comparative study to concurrently evaluate multiple normalization methods for DBS metabolomics which will serve as a robust methodological platform for future environmental epidemiological studies.