Random terrain generation is the procedural creation of a set of data that represents closely a believable landscape. Common techniques of achieving this include the use of fractals and noise. Such techniques usually require a large volume of memory, as the geometry of the terrain needs to be calculated and stored at run time. Given the limited memory available on mobile devices, such as mobile telephones, the storage of the data required to represent massive terrains can be difficult. In this paper, we propose a novel method of storing terrain data on devices with limited memory. This method involves placing pre-computed blocks of terrain, known as terrain tiles, together in a psuedo-random manner as governed by a noise function known as Perlin noise. This allows large amounts of terrain data to be represented while still giving the appearance of a randomly generated terrain. Traditionally, Perlin noise is used in the procedural generation of textures and the modeling of naturally occurring phenomena. Using Perlin noise, only a subset of the overall data generated by the function needs to be stored at any one time. The approach outlined in this paper associates a tile of terrain with each value generated by the Perlin Noise function, meaning that only a subsection of the total terrain is stored in memory at any one time. We show how this process can be executed in real time on a resource limited device known as the Game Boy Advance, and also illustrate a significant reduction in the memory requirements of terrain storage when compared with traditional methods.