You have an excellent answer from Remi.b already but I just wanted to add/emphasise this (because there is always more than one way of explaining something and IMO the site benefits from having many answers to the questions)...
The tree we construct does not necessarily accurately reflect what happened in evolution. If B & C evolved from A, and D & E came from B, we would create this tree if we measured using the correct indicator. But the methods we have are not perfect. The first evolutionary trees were based on morphological descriptions etc. and clearly some of the classifications were going to be wrong. These days we use molecular methods, which are probably more accurate but could also be wrong some times. For example if we based our phylogeny on one SNP variant we could have some idea about the phylogeny between a few species, but if we based it on millions of SNPs we would have a much better idea - as technology & models improve that is becoming more realistic. The key point here being there is a difference between the trees we can draw from evidence, and the real evolutionary tree.
If, for example, D evolved from B but was more genetically similar to C it could be due to convergent evolution (if the same mutations are selected for in C and D but neutrally or negatively in B). This is because, if there is a shared standing genetic variation at the point of speciation or the same mutations occur after speciation, and then C & D are imposed under similar selection and B is not, the allele favored in that environment will spread in C & D but not B (it may drift in B).
Imagine this with humans, where populations split off around the globe. Two populations spread from Africa (A) in to Southern Europe (SE) and Western Asia (WA). Population WA then derived two more populations - Central Asian (CA) and Eastern European (EE). If the Eastern European population developed a habit for drinking cow milk, and that information passed to the Southern Europeans, then selection would favor the ability to digest lactose in those two populations. If we then used the lactase persistence SNP (which had standing genetic variation at the time of divergence) to derive a phylogeny the populations SE and EE would appear related because the allele for lactase persistence is common, but in A WA & CA it would be rare (neutrally evolved). In this case the phylogeny we construct would be different to the true phylogenetic tree.
Basically if there is genetic variation at the point of divergence, or convergent selection, or any gene flow occurs after divergence it makes assessing the phylogeny more difficult and prone to error.