XmlNodeList is an abstract class that represents an ordered collection of XML nodes and it can be returned by XmlNode.ChildNodes, XmlNode.SelectNodes and XmlElement.GetElementsByTagName. The documentation here clearly states that XmlNodeList is a “live” collection so that changes to the children of the node object that it was created from are immediately reflected in the nodes returned by the XmlNodeList properties and methods. The interesting thing is that if you retrieve XmlNodeList using ChildNodes or GetElementsByTagName - it is indeed a live collection. Remove or insert a node and bingo - XmlNodeList is up to date automatically. However, when you retrieve XmlNodelist from XmlNode.SelectNodes, - the collection is “semi-live“. That is the nodes are cached as soon as you access them and if say, you access Count or the last node - all the nodes will be cached internally and changes to the document will not show through this XmlNodeList because the cache is never reset. So it’s a nice example of a not so nice thing: XmlNodeList is an interface and therefore a contract that is broken by different XmlNodeList implementations. The point is: This code is NOT supposed to work because the collection is live: XmlNodeList shapes = elem.GetElementsByTagName(“Shape“); for (int i = 0; i < shapes.Count; i++) shapes[i].ParentNode.RemoveChild(shapes[i]); However, this code coincidentally (nastily) works: XmlNodeList shapes = elem.SelectNodes(“.//Shape“); for (int i = 0; i < shapes.Count; i++) shapes[i].ParentNode.RemoveChild(shapes[i]); I was looking at XmlNodeList in Rotor source and from what I can guess, .NET probably implements it the same way although I did not bother confirming myself.