The 4096 Processor DAP has been installed at Queen Mary College for over one year. A wide range of problems have been programmed for the DAP and the performance of the system exceeds the original estimates. The DAP is programmed using DAP-FORTRAN a FORTRAN dialect with many parallel programming extensions. Experiments have shown that few problems benefit from using the low level assembler language (APAL) and so assembly level programming is now strongly discouraged. After initial resistance to change, sceptics rapidly fall in love with the increased power of a parallel FORTRAN and find beneficial the disciplines involved in a re-evaluation, at high level, of their problems. Comparison with results of calculations on serial machines shows that the DAP produces the maximum benefit in problems dominated by conditional operations (i.e. FORTRAN IF type statements). This result suggests that the concepts of 'speed-up' and 'efficiency' often used in parallel algorithm theories are dangerous when applied to the DAP class of architectures. There is no single measure of the absolute performance of total computer systems and one can only measure the performance of one system in relation to another. Relative to familiar serial computers the performance of the DAP depends strongly on the problem type and the size of the problem. The MEGAFLOP unit proves to be totally meaningless when applied to the DAP (the theoretical maximum performance is 0.0 MEGAFLOPS!) but when properly programmed the DAP can, for a large rar.~eof scientific problems out-perform any other computer system.