puzzling results when contrasting conditions in 3 x 2 design

Hi everyone,
I'm encountering something puzzling when analyzing my data. My data consists out of 2 independent variables and one dependent variables. When analyzing my data in a 3 (conditions: A, B, C) x 2 (Conditions: D, E) between subjects design, using Univariate Analysis of Variance in spss assigning 2 fixed factors. Everything works out as it should be.*
One of the three conditions is the control condition. What I really want to check is whether there is an interaction of A and B with D and E when predicting the dependent variable. My first approach was selecting only cases in condition A and B and do the 2 x 2 analysis. This analysis works out fine as well. Than I thought the analysis would look nicer when using all participants and test*A and B*contrasts. When testing simple*contrasts*between A and B the data looks less*nicely,*how could that be? I thought that using more participants makes the power and*consequently*the*chance*of obtaining significant results bigger. Why is*this*not the*cases when using the contrast option in the Univariate Analysis of Variance in spss, but is the case when only selecting the two conditions of interest?
Any advice would be much appreciated.