In this meta-analysis of 34 clinical studies on attachment the hypothesis is tested that maternal problems such as mental illness lead to more deviating attachment classification distributions than child problems such as deafness. A correspondence analysis on 21 North American studies with normal subjects produced a baseline against which the clinical samples could be evaluated. Separate analyses were carried out on studies containing the traditional A, B, C classifications and on studies that also included the recently discovered D or A/C category. Results show that groups with a primary identification of maternal problems show attachment classification distributions highly divergent from the normal distributions, whereas groups with a primary identification of child problems show distributions that are similar to the distributions of normal samples. The introduction of the D or A/C classifications (about 15% in normal samples) reveals an overrepresentation of D or A/C in the child problem groups, but the resulting distribution still is much closer to the normal distributions compared to the samples with maternal problems. In clinical samples, the mother appears to play a more important role than the child in shaping the quality of the infant-mother attachment relationship.