From the documentation the following rules of thumb may be gleaned: x) keep heavily acessed fields (key and index fields) at the front of the record x) keep large, variable sized fields (clob and blob) at the end of the record x) sometimes it pays to split off clob/blob information from the main record The above does not save space, but saves tons of effort when decoding the compressed records, as acccessing field n requires decoding of fields 1..n-1 even if they are not used.