All programs currently running on serial computers require some degree of modification when moved to parallel processors. This is true whether the architectural parallelism is manifested at the instruction level, such as in array processors or the CRAY, or achieved via multiple processors, as is the case in the MIDAS system. In either case the degree to which the program exploits the architecture can significantly affect the processing speed. Some guidelines for application programming for the MIDAS system are discussed. Important programming considerations include the separation of serial and parallel elements of a program (such as program initialization), data input mechanisms (including hardware preprocessing), and output mechanisms. Comparisons of code written for standard serial machines to the same code modified for MIDAS will be examined and performance results discussed.