This paper describes a series of tests on the accuracy of DEMs. The effects of 4 factors on DEM accuracy have been tested, i.e. the accuracy, density of source data, the characteristics of terrain surface and the modelling approaches. A large area covered by two 1:10,000 maps were selected for testing. The terrain types range from flat, hilly to mountainous. Various sources (photographs and contour maps) are used and measurement methods employed. From test results, the followings have been found: (a) Source data measured manually from analytical plotters is most reliable as automated measurement using image matching could generate systematic errors; (b) The accuracy of DEMs decreases (or RMSE becomes larger) with an increase in sampling interval and the relationship is linear; (c) The inclusion of feature points and lines improves the accuracy of DEMs significantly, especially when the terrain is rough; (d) Generally speaking, the accuracy of DEMs decreases with an increase in relief, however, it is not always the case and the best results may be obtained in hilly areas; (e) The direct modelling from originally measured data to form triangular network will yield better results than the indirect modelling via a random-to-grid interpolation to form grid network. The difference could be significant if the terrain is rough. From these conclusions, some advice on DEM production could be made as follows: (a) When using automated photogrammetric system for data acquisition, editing by experienced operators should be considered; (b) In hilly areas, photogrammetric contouring can be most efficient methods for DEM data acquisition if analytical plotters are used; (c) Feature points should always be measured and kept in order to reduce the data volume while retaining the fidelity of the DEM; and (d) When terrain surface is rough, triangulation-based methods are recommended.