Only a small number of protein homologues cause the majority of allergies. There is no consensus structure or other obvious common denominator discriminating the few proteins that are allergens from thousands of other, non-allergenic proteins. By database sequence homology searching, we here show that to date known allergen sequences have no or few bacterial homologues, in contrast to randomly selected control protein sequences. This finding suggests a novel common denominator for allergens of potential use for allergen prediction programs. A possible interpretation of this finding is that allergens are proteins which are exposed to the immune system and which lack bacterial homologues. This interpretation is discussed in relation to the many observations that allergies coincide with a delayed establishment of infant gut flora. (c) 2007 Elsevier Ltd. All rights reserved.