The manufacturing of high-quality chrome masks used in the display industry for the manufacturing of liquid crystals, organic light emission diodes and other display devices would not be possible without high-precision large-area metrology. In contrast to the semiconductor industry where 6′ masks are most common, the quartz glass masks for the manufacturing of large area TVs can have sizes of up to 1.6 × 1.8 m2. Besides the large area, there are demands of sub-micrometer accuracy in ‘registration’, i.e. absolute dimensional measurements and nanometer requirements for ‘overlay’, i.e. repeatability. The technique for making such precise measurements on large masks is one of the most challenging tasks in dimensional metrology today. This paper presents a new approach to two-dimensional (2D) ultra-precision measurements based on random sampling. The technique was recently presented for ultra-precise one-dimensional (1D) measurement. The 1D method relies on timing the scanning of a focused laser beam 200 µm in the Y-direction from an interferometrically determined reference position. This microsweep is controlled by an acousto-optical deflector. By letting the microsweep scan from random X-positions, we can build XY-recordings through a time-to-space conversion that gives very precise maps of the feature edges of the masks. The method differs a lot from ordinary image processing methods using CCD or CMOS sensors for capturing images in the spatial domain. We use events grabbed by a single detector in the time domain in both the X- and Y-directions. After a simple scaling, we get precise and repeatable spatial information. Thanks to the extremely linear microsweep and its precise power control, spatial and intensity distortions, common in ordinary image processing systems using 2D optics and 2D sensors, can be practically eliminated. Our 2D method has proved to give a standard deviation in repeatability of less than 4 nm (1σ) in both the X- and Y-directions over an area of approximately 0.8 × 0.8 m2. Only feature edges are recorded, so all irrelevant information in areas containing constant intensity are filtered out already by the hardware. This relaxes the demands and complexity of the data channel dramatically compared to conventional imaging systems.