This can be done in various ways depending on what equipment you have access to. The most accurate is with an oscilloscope and signal generator. Another method is to use JavaTC the software design program, which will give a good estimate of the resonant frequency and where you need to tap the primary coil.
In both cases you can get the best final tuning point by testing different tap points close to the estimated point, to get the best result.
It is also always a good idea to initially run at a low power, as if the coil is badly out of tune and ran at full power, you could cause damage, either to the secondary coil or the capacitors.

If you have not used JavaTC then now is a good time to get it and use on the coil you have already built, otherwise assuming you do not have an oscilloscope, you could calculate the resonant frequency yourself. See Here or Here for the necessary formula's.

The basic procedure is either measure or calculate the resonant frequency of the secondary first, and then use this value to work out the correct point to tap the primary. The result should be that the tapped inductance of the primary, when combined with the capacitor (MMC) value, will result in a resonant primary frequency that is around 5% to 10% lower than that of the secondary.

Strictly speaking the primary frequency should match the secondary frequency, but when a streamer forms on the toroid it adds capacitance to the circuit and causes the resonant frequency to drop, something that is often overlooked by a lot of people. The amount that it drops will vary depending on how long the streamer is, and where, or indeed if, the streamer strikes anything.

An approximation of the amount of frequency drop can be obtained by trailing thin pieces of wire from the toroid to nearby objects, and then remeasuring the resonant frequency of the secondary circuit again. These simulated streamers should only be the length that you know the coil is capable of.
By trying different lengths and positions you can then come up with an average frequency drop to aim for. You then tune the primary to this new frequency.
It will mean of course with this method that the primary is initially slightly out of tune before a streamer forms, but it then comes fully into tune when a streamer is present.
This initial out of tune period does not seem to affect things, but the fact that it comes into tune when a streamer forms, makes for longer streamers or sparks overall. Unfortunately you can only use this method to make allowance for the effect of streamers if you are using an oscilloscope and a signal generator to tune your coil though.



(Above) Tuning the Primary
The 10K** (approx) resistor is very important so don't forget it. Without it the scope is just sitting across the generator's output, so it will only see that output voltage and hide that of the coil's resonant rise.
** 10K is the best value for my own signal generator, you may need a different values.






(Above) Tuning the Secondary



The instruments are connected as shown above (the wire simulating a streamer is optional). The probe must not swing or move around, otherwise it will upset the readings. Set the signal generator to an initial frequency that is within about 20 to 30 KHz below the expected frequency. Then gradually increase the frequency and watch the oscilloscope display.
At some point the voltage that the 'scope is measuring will sharply increase, then abruptly decrease. Just how suddenly this occurs is affected by the coarseness of the tuning and the coils own characteristics. I modified my own generator so that I had quiet fine tuning in the range of 50 to 150 Khz (other coils may well resonant outside this range).
You will most likly find that there are several points where the voltage increases, but there will always be a dominant one that is more than others. This will be located closest to the expected frequency, whilst the other 'false ones' may be fractions of the expected frequency (1¼ , 1½ etc) This can be seen in my YouTube video clip below:

Some people say the primary should ideally be tuned without the secondary in place, but I feel that as the secondary is in place when the coil is running, then any effect its proximity has, should also be present when you are tuning. The secondary circuit (with toroid in place)can also be tuned while it is mounted on the coil. If though you have previously been tuning the primary, you must remember to remove the shorting link across the spark gap before doing the secondary!


The results below (from a 4 inch coil) attempt to show the simulated effect that streamers have. Secondary Coil Results:

Freq =	148.1 Khz	(No streamer)
Freq =	135.9 Khz	(One simulated streamer)
Freq =	127.8 Khz	(Two simulated streamers)

The last but one reading of 135.9 Khz was the frequency that I decided to tune the primary too, as this represented the most likely scenario with the coil running.
JavaTC does not allow for this method of tuning the primary deliberately low unfortunately, as if you choose its option to manually tune the primary for you, it will only give you a tapping point that actually matches the secondary frequency.
Possibly JavaTC's only shortfall as far I can see.

Now here comes the troublesome part On the primary for my 8 inch  'Phoenix' coil, the lead from the MMC to the primary was a bit long and greatly affected the results. If the lead was routed in one direction it would give a different frequency than if routed in another direction to exactly the same point. This is something that you must check as it can catch a lot of people out if your not expecting it. The answer of course is to shorten the lead so it runs direct, once you know where you want the tapping point to be.
This is similar to the concept of using 'off-axis' tuning where you have a separate variable inductor in series with the main primary, but in my case it is unintentional and unwanted.


A Circuit to view the resonance.



This circuit is attributed to Antonio Carlos M. de Queiroz and can be found along with a lot of other very interesting stuff on his site Here.. Antonio suggested using a 1 ohm resistor across the spark gap but I found I needed a 0.5 ohm to be able to see the correct waveform. The frequency from the signal generator must be square wave and it needs to around 1 to 4 Khz for best results.