Conflict early warning remains an important but elusive goal in Liberia. If outbreaks of violence could be predicted before they occur, early responders could focus their energies and scarce resources on the highest-risk communities. Is such a goal realistic? Early warning requires a simple system for generating reliable predictions—a system that is not only accurate, but is also consistent over time and across counties and communities. This is a difficult, maybe impossible, task. This report describes results from a two-year study that suggest prediction may be more promising than we initially expected. The report uses fine-grained quantitative data from a survey of 247 rural Liberian towns and villages to assess whether statistical analysis can be used to predict conflict over time. Models built on fewer than 10 risk factors measured in 2008 accurately predict up to 75% of all conflicts two years later. These models are flexible, and can be trained not only to maximize accuracy, but to minimize “false negatives” as well. These models accurately predict 40 to 70 percent of all incidents of conflict (“true positives”), with three to five false alarms (or “false positives”) for each correctly predicted incident. The models are also able to distill a small subset of robust predictors from a long and unwieldy list of potential risk factors. Out of 60 potential risk factors, each model identifies 5 to 10 correlates that are especially powerful in predicting the onset of conflict, and that government and NGOs might use to target high-risk communities. Finally, the report concludes with recommendations for advancing the goal of conflict early warning in Liberia. First, the early warning/early response (EWER) community should focus less exclusively on analysis of past conflicts and more on anticipation of future ones. Second, researchers and NGOs should standardize their indicators for conflict and risk. Third, to test whether these (or alternative) models are consistent outside our sample, the report suggests three possible next steps for data-driven EWER.