AbstractIn the search of data from a database using a key, it is desirable that, even if the input key or the key in the database is in error, and there is no exactly matched key, still the system lists similar keys or retrieve the closest key.This paper discusses the organization and manipulation of the file, where the keys similar to the input key are listed, or the closest key is retrieved. The basic idea is to combine the high efficiency of the B‐tree for the retrieve/insert/delete of the key, and the ability of the hierarchical file based on the class name expression for the similar key search. By such an elaboration, it is possible to realize the search of similar keys, which cannot be executed by the B‐tree, or insert/delete of the key, which is not considered in the hierarchical file based on the class name expression.An experiment is made by splitting the characters into two classes, i.e.,atomandntoz, and using 16,561 English words of length 6 to 10. The number of read‐outs from the secondary memory in the retrieve/insert/delete of the key is approximately 3, and the number of write‐ins into the secondary memory in insert/delete is approximately 1. The search for the similar keys requires approximately 7.3 to 8.5 times larger read‐ins compared to the search of the exact key. The retrieval rate is approximately 95 to 99 percent for a single error (substitution/insertion/missing). The efficiency of the use of the memory is approximately 7