The behavior of the occupants is one major factor determining the outcome of building evacuations. Two different occupants may make very different decisions in similar situations. Evacuation simulation models should consider this factor and enable simulations with different behavioral scenarios. We present how three different behavioral types have been implemented in the agent based evacuation model FDS+Evac. The considered types are "conservative", "active", and "herding" agents and they differ in the way they select their target exits. Conservative agents prefer familiar routes, active agents actively observe their environment to find the fastest exit route, and herding agents tend to follow others. The factors affecting the exit selections include the distance to the exit, the behavior of the other agents, visibility and familiarity of the exits, and the smokiness along the exit routes. Some verification simulations are made and the presented model is seen to work as intended. A larger scale evacuation scenario is analyzed to illustrate the effect of the agent types on the outcome of evacuations. The analysis shows that the presence of active agents decreases the average egress time even if there is a large number of herding agents present. Active agents discover the less used, faster exit routes, like emergency exits, and lead the following herding agents also to these routes. This results in more even exit usage and speeds up the whole evacuation.