The standard Hough transform could be used to solve recognition tasks in range data, if it were not for the high dimensional parameter spaces these tasks require. This paper describes two alternative methods which apply search techniques at multiple resolutions in order to find sets of parameters which fit the data best. Both methods are as robust as but much faster than the standard Hough transform. The first method, called recursive lattice search, employs a data structure similar to the quad-tree or oct-tree in order to attain its efficiency. The second method, called resolution hill climbing, finds a trail of hypothesized parameter sets, each of which fits the data at a higher resolution than the one preceding it. An example two dimensional recognition task and an example three dimensional recognition task are considered both without and with gradient information. Finally, a more abstract type of feature, the reflection symmetry, is shown to be recognizable by a modification of the first method. Some results of an implementation of the two methods are described.