The use of flexible robots can be easily justified in two main cases: (1) when the weight of the robot has to be minimized and (2) when collisions between the robot and the environment are foreseen, since a flexible, lightweight robot implies less impact energy. The position control of these robots has already been analyzed in previous communications. However, the second of these cases justifying the use of flexible robots requires further consideration, leading to the development of a force controller. In most up to date analysis the force control is studied beginning from a known contact point at a given collision time. In a more realistic approach, however, an accurate detection of the collision would be needed prior to dealing with the force control. After developing a reliable position controller for a three-degree-of-freedom flexible robot using strain gauges placed over the robot structure, in this paper we deal with the possibility of carrying out the collision detection for the same prototype. This has been easily achieved by analyzing some estimated signals such as tip position and tip velocity. However, a complete analysis of the information obtained with the sensors has been required to obtain those estimates and a signal processing scheme had to be devised for a previous filtering of the original signals from the sensors (encoders and strain gauges). This work has been carried out as a first step towards the position/force control. Experimental results on a three-degree-offreedom flexible arm prototype are presented to verify how well this method performs.