Has anyone ever seen the distributions that result when plotting percent voiding? (Typically a BGA will have enough balls that you can get oogles of data.) I have seen this and seem to be noticing a pattern. First, they are not all normal, so your traditional CpK analysis won't apply (you can do a non-normal process capability study). The best balls seem to have an exponential distribution. Ppm's above 15% voids are single digits. Next you will find a Weibull distribution skewed left, also with one to three digit ppm's above 15% voiding. Last, you will find a normal distribution to which you can apply the traditional CpK analysis. Any other distributions thereafter will have generally unacceptable results. Hence, because I see that the distributions are usually not normal, one should not look at the mean percent voiding as significant. I think one ought to pay more attention to either the median or what the amount of ppm's are above a certain upper percent voiding limit. But what is acceptable ppm then? Assume that you have a sample which shows the rest of your production has 2000 ppm above 15%. If you have 10,000 balls in that board, then 100 of these boards (a not ambitious production run) will equal a million balls. Therefore, in those 100 boards you have 2000 balls with more than 15% voiding! In other words there are at least 20 balls per board that have problems! Though 20 out of 10,000 is just 0.2% that small number does not "seem" to me to be small enough. Hence, I think that in these matters you really need single or double digit ppm's above your upper percent void limit. (Admittedly the IPC limit is that no more 20 or 25% voiding is good, I can't remember which, but I was being strict and set the limit at 15%.) Any thoughts?