The main research objective of this study is to gain more insight into the uncertainties that are occurring in maintenance projects. A lot of research has already been performed on the subject of budget overruns in large scale new development infrastructure projects, however, information on budget overruns in maintenance projects is lacking. Uncertainties are an important contribution to the cost increase of (infrastructure) projects. During a project unforeseen events are always occurring. This is taken into account by adding an item ‘unforeseen’ to the cost estimate. The actual occurring unforeseen costs can be split up into three categories: unforeseen costs in the execution phase, contractors bid uncertainty and ‘other unforeseen’ due to, amongst others, further specification of the project. It appears that in new-to-build projects the unforeseen in the execution phase (quantified by additional works) can be as high as 25 to 30 percent of the total project costs. In maintenance projects this is a lot lower: a maximum of five percent. This has to do with the repetitive character of maintenance works, the often limited project size, the possibility for detailed inspections and the fact that the works often take place in already stirred ground. Though, maintenance projects are characterized by a relative large uncertainty between the initiation phase and the start of the execution. This is because in maintenance projects one never starts with a ‘green-field’ situation. In other words, the initial state is often unclear in the early phase of the project, and this is especially the case for unique, technical complex projects, such as control system projects. Because this report was initiated by the Airfield Maintenance Services (AMS) division of Amsterdam Schiphol Airport, applying the information found on the basis of the main research objective is a second goal of this thesis. It is in the ambition of AMS to be reliable and predictable. Not only when it comes to the management of its assets, but also when the financial performances are concerned. Also within the rest of the organization accurate cost estimates are seen as an important means towards more efficient and effective cost control. This enables them to make better informed decisions. Therefore, the second research objective of this study is to improve the early-phase cost estimation process of maintenance projects and to find out whether probabilistic cost analysis techniques could be of value in this process. The scope of this study is limited to the CAPEX projects; the large renovation projects. The current practice in the cost estimation process at SCHIPHOL GROUP is that every year in the first quartile a business plan for the upcoming five years is made based on the company’s strategic goals. The input for this business plan is the budget estimates that are provided by the maintenance managers of AMS. At this moment uniformity is lacking in these estimates. Moreover, when the estimates are made the scope of the project is often still very unclear. Based on this business plan, after further specification, the project budgets are determined in the fourth quartile. The annual budget for the next year is the project budgets combined. From a comparison between the business plan, the annual budget and the actual expenditures in the projects it appeared that the latter are significantly lower. This can be mainly explained by the fact that throughout the year projects have been removed from the business plan scope. However, a statistical analysis of the projects itself shows that on an average the expenditures were 7% higher (with a standard deviation of 34%) than estimated in the business plan. This can be further specified per project type. Because the project budget estimates were made in a later stage, they are more in line with the actual costs. What also showed from this analysis is that smaller projects show a larger variation in the nominal unforeseen than larger projects. This could indicate that in maintenance projects larger projects can be seen as a cluster of smaller projects with relatively little overlap between them, thus acting as a portfolio where overruns in one part of the project are compensated by underruns in another. The cost estimation process can in first instance be improved by creating uniform estimates in combination with a clear definition of the scope. When, next to this, the costs are booked into the accounting software system in a corresponding manner, a database can be built with very usable information. At first this information can be used to gain insight into the size of the cost items. If needed, it will then be able to make decisions accordingly. Further this can be used as a starting point for new estimates, which will make them more reliable. In the second place it is recommended to do the estimate in a probabilistic manner. This means that the total estimated amount is represented by a probability function with an average value and a standard deviation. With this a probability of exceeding the chosen project budget is introduced. In this way one recognizes the uncertainty of an estimate and this information can be used to determine the budget. There are different methods by which a probabilistic cost estimate can be made. In this thesis it is recommended to do this statistically (with the use of historical data) and not Bayesian. This is related to the issue that it appears to be difficult for cost estimators to estimate extreme values, uncertainties and risks. It should be noted that when historical data is not present, the Bayesian method can be used until enough statistical information is gathered. For the determination of the project and the annual budgets the following is proposed. When the annual budget is applied for one can take the uncertainty of the estimates into consideration. This becomes relatively smaller when the projects are bundled in a portfolio and regarded as a whole. The annual budget can be set to a value equal to the sum of the project cost averages plus a value of k times the standard deviation. Subsequently the projects can be given a budget with a probability of exceedance of 0.5 or even higher. Next to this a contingency fund with a size of the earlier mentioned k times ? can be kept at the management level of AMS. If it seems that the projects will be more expensive than the set budget the project manager can request for extra budget from this contingency fund, but only on the basis of solid argumentation. It should be noted that it is not in everyone’s interest to make the estimates more accurate and transparent. Some stakeholders could benefit from a large budget and more freedom when it comes to the allocation of financial resources. This could lead to strategic behavior. Budget slack is an example of this. This means that one applies for a larger budget to make it easier to reach a target. Also there is the MAIMS (money allocated is money spent) principle, which means that one is inclined to fully spent a given budget, even though this is not always necessary for the originally set scope and quality of the project. By using the in this thesis proposed estimation method and budgeting process, these effects can be diminished, by increasing the transparency and starting off with tighter budgets for the projects. Besides, it is regarded in the best interest of the entire organization when transparency and reliability of the estimates is increased, such that more efficient and effective cost control is possible. Finally, the estimation process of Schiphol AMS can be improved by starting the projects more early. When there is already a preliminary design of a project before the budget is set in the final business plan, a part of the uncertainty in the scope and in the further specification can be reduced. Information of the project organization within SCHIPHOL GROUP and the contractor can serve as valuable information here. This does not necessarily have to cost extra money and time, since a part of the engineering of the project that needs to be done anyway is only moved forward in time. On top of that there is added value in the fact that more reliable estimates can be delivered.