Finding similar objects is an essential operation, which is widely required in many applications such as data cleaning, near duplicate detection, query relaxation, and spell checking. To quantify similarity between two objects, existing similarity functions are mainly based on the number of common elements between the objects, where an object is represented by a set of comparable elements. However, those similarity functions does not find a similar pair of two objects, when the objects does not share a common elements even though elements from the objects are very similar. To overcome this problem of existing similarity functions, we propose a record similarity function, which consider similarities among elements in the objects. Considering many important data objects are represented in textual form, we assume that an element in an object is a string and the similarity between elements are measured by the edit distance. Naturally, the proposed function can detect a similar pair of objects consisting of similar elements. A drawback of the proposed function is that the complexity of the record similarity is higher than existing similarity functions. Given a database containing many objects, it is computationally prohibited to find objects similar to a given query object by comparing the record similarity between the query and each object in the database. To efficiently find similar objects, we also propose novel filtering and verification techniques. Exploiting a given record similarity threshold, the proposed filtering techniques effectively reduce the number of candidate objects and the proposed verification techniques further reduce the final candidate objects by refining initially generated candidates. Through an experimental study, we show the record similarity function precisely find similar objects compared with the existing similarity function. We also show proposed filtering and verification techniques effectively reduce the number of candidate objects that require the full record similarity computation, and thus, they enable efficient processing of similarity search and join.