It is now commonly accepted that the unit disk graph used to model the physical layer in wireless networks does not reflect real radio transmissions, and that a more realistic model should be considered for experimental simulations. Previous work on realistic scenarios focused on unicast, while broadcast requirements are fundamentally different and cannot be derived from the unicast case. Therefore, broadcast protocols must be adapted in order to still be efficient under realistic assumptions. In this paper, we study the well-known multipoint relay broadcast protocol (MPR). In the latter, each node has to choose a set of 1-hop neighbors to act as relays in order to cover the whole 2-hop neighborhood. We give experimental results showing that the original strategy used to select these multipoint relays does not suit to a realistic model. Based on these results, we propose new selection strategies solely based on link quality. One of the key aspects of our solutions is that they do not require any additional hardware and may be implemented at the application layer, which is particularly relevant to the context of ad hoc and sensor networks where energy savings are mandatory. We finally provide new experimental results that demonstrate the superiority of our strategies under realistic physical assumptions.