For targets that support variable length instructions, then there will be multiple decode tables to parse through and memory may have to be read multiple times before a successful decode occurs.

One approach to this which is be suitable in most cases is to start with the smallest instructions, reading the smallest amount of memory possible and then using the appropriate decode table.

If that fails to match an instruction, then more memory should be read, trying the next smallest instruction table, repeating until all tables have been tested. If no match has then been made an error code should be returned.