Fifteen studies fulfilled the inclusion criteria (21–35) and were tabulated. These papers differed so markedly in study populations (Table 1), types of intervention, trial design, duration (Table 2) and outcome assessment methods (Table 3) as to withstand systematization or meta-analysis. Definition of atopic dermatitis was defined according to clinical criteria in all studies, sometimes with a specific reference (e.g. Hanifin's criteria). In three studies ‘severe’ eczema was considered an inclusion criterion whereas three other considered ‘refractory’ eczema as sole entry criterion. A selection bias appeared in all studies as children with eczema were referred for assessment to either dermatological or paediatric allergy clinics. Their caseloads are not representative of the whole population of infants and children with AEDS. The disparity between entry criteria was matched by differences in sample size. In total, 823 children were assessed; 142 in caseloads recruited on the ‘refractory’ disease criterion (25,28,30), 137 in studies concerned with ‘severe’ disease (22,29,33) and 544 in studies which did not specify the severity of symptoms for enrolment (21,23,24,26,27,29,31,32) (Table 1). There was no uniformity in the exclusion/trial of foods as part of the dietary intervention. Two studies eliminated cow's milk and egg (21,23), two prohibited only eggs (27,34), four studies substituted formulae [either elemental (25,31,36) or cow's milk hydrolysates (31,32,36) to their patients while two others excluded cow's milk and eggs from the diet of the nursing mother (24,27)]. The elimination of cow's milk and/or eggs (22), of ‘suspect’ foods (26), of three (28), or six (29) foods has also been recommended as well as an elemental diet based on a small number of foods (30,33) (Table 2). The main limitation of published trials in this field is the lack of blindedness. One study (24) compared the effect of an exclusion diet with a placebo under double-blinded conditions but only in a subset of patients. Another study (32) carried out a blinded comparison of probiotic supplementation, but the allergy arm of the study was open. In mitigation, it must be said that finding suitable placebo is difficult. An amino acid elemental formula could substitute for cow's milk, but the two studies assessing its use against a whey hydrolysate are open studies (31,35) (Table 2). Well-controlled, prospective assessment of cow's milk restriction is ruled out by ethical concerns. Other confounding factors present in all these studies are not always acknowledged. The most conspicuous is a method for evaluating eczema for which a consensus is yet to be reached, although specific scoring systems have been devised (36,37). Thus studies rely on home-made scores, but some are difficult to use (38) and evidence or consensus on which is the best are still lacking (39,40). Another confounder is the method used for diagnosing food allergy in these children. In some studies, children are sensitized to one or two foods but do not undergo formal demonstration of food allergy by oral provocation tests (22,25,34). In other studies children are sensitized by multiple foods (33) or are described as suffering from atopic dermatitis but without specifically referring to the foods they are sensitized by or allergic to. Only two studies enjoy the ideal situation when an allergic reaction to the study food is confirmed at double-blind, placebo-controlled food challenge (DBPCFC) (31,35). A third factor is the problem of the evaluation of compliance with dietary advice. Dietary restrictions are increasingly hard to enforce among older children and in children undergoing stringent dietetic prescriptions the phenomenon ought to be taken into account. Of further limitation are the high drop-out rates [reported as 44 (21), 25 (23) and 41% (30)]. All told, 112 out of 823 children (or 13.6%) were lost to follow-up in the studies considered. A further confounder of study interpretation is the concomitant administration of nondietetic therapeutic intervention, e.g. pharmacological treatment (Table 1) and environmental measures (25,29). The limited duration of many of the studies is also a drawback: most studies investigated only short- to medium-term effects. Three studies evaluated patients 6 (35), 9 (31) and 12 (25) months into diet therapy, each with favourable conclusions. In a disease with marked seasonal variations, it can be impossible to draw conclusions from studies of 2, 4 (21,28,32–34) or 6 weeks (29,30). Finally, many of these studies have a lack of statistical power because of poor study design and only in one study of cow's milk allergy in children (31) was the effect of a substitute assessed against placebo use. Case-controlled cross-over studies can be fitted to logistic regression models but these studies may not have sufficient statistical power to evaluate the multiple outcomes of an exclusion diet on AEDS (14). A few studies focused on the risks of exclusion diets in AEDS. We know that children with eczema show a reduced pattern of growth during the first year of life (41) and this can reflect a nutritional disorder because of food allergy. These disorders may be aggravated by inappropriate diet, sometimes with dramatic effects (42). Another case in which eczema can develop in the absence of food allergy even in little children is phenylketonuria (PKU), when the appropriate metabolic diet can prevent both atopic eczema and the allergic sensitization to foods (43). A study showing that feeding an amino acid formula resulted in improved growth compared with feeding an eHF suggests that the more anallergic are the diet substitutes, the more adequate is their nutritional value (35). Thus, a full nutritional history has to be taken when exclusion diets are planned for children with AEDS.