Wind speed in a building, windows and angles as factors: is ANOVA appropriate? EDIT
Is an ANOVA appropriate for rhe following?
I have a physical cuboidal building in an open field with a single window (S) or 2 windows (X). 

1. Single window cube (S) : I measure a variable (air speed) at 10,000 points inside this cube (blue plane) and compute the average. I then turn the cube a few degrees with respect to the oncoming wind (angle) and repeat the experiment.
2. 2 window cube (X): Repeat with same angles.
I'd like to know if there is a way of identifying which factor (angle or window set-up) is more important for varying the air speed inside?
I now have a table of values:
z= speed angle window
    15 0 S
    20 0 S
    25 0 X
    04 0 X
    55 15 S
    16 15 S
    15 15 X
    20 15 X

...
In R then I do:
ACH.aova <- aov(speed~ window*angle, data=z)
summary(ACH.aova)
TukeyHSD(ACH.aova)
plot(TukeyHSD(ACH.aova, "angle"))

But I'm not sure if this is the best method or how to interpret the results... Any help would be much appreciated,
 A: 
But I'm not sure if this is the best method or how to interpret the results... Any help would be much appreciated,

On interpreting results: Please specify if there is something in particular you are unsure of. When it comes to ANOVA: The F-value describes whether the variation between groups is the same as within groups. E.g.:
Where the null hypothesis states:
H0 = the two groups are equal
And the Alternative hypothesis states:
HA = the two groups are not equal
If the null hypothesis (H0=the two groups are equal) is true, than F should be near 1, deviating from it only by chance. The P-value is the probability that the variation seen between the groups is due to chance.  
Whether this is the best method: See below.

I've just realised my air speeds on the blue plane are not normally distributed... This is a problem for ANOVA isn't it? I guess I am looking for a two way factorial test for non normal output variable. What do you think?

If your data was normally distributed, you could have run a paired T-test. Since you have too little data to assume a normal distribution, and your data doesn't follow a normal distribution, you need to use a non-parametric (does not require a normal distribution) method. Also, the F-test is highly sensitive to non-normality. The Mann-Whitney U test is a non-parametric alternative to t-test or ANOVA for two groups. 
Here is a page from Quick-R with non-parametric testing of group diffferences:
http://www.statmethods.net/stats/nonparametric.html
