How to measure resistance of DI water It is known that good DI water have resistance ~20 MΩ·cm
But how can I measure that? Using good vanilla ohmmeter (with 2000 MΩ range) showed crazy results (too low, not much dependent from distance between probes).
I need that to compare DI water from 2 sources.
 A: Resistance (that's what your meter reads) is related to resistivity (that's the 20M Ohm x cm term) via the geometry of the problem.  If you have a body of the measured substance with UNIFORM CROSS SECTION (doesn't matter what shape) between two parallel and highly conductive plate electrodes (one plate on each end of the measured body, and in full contact with it), then:
Resistance = Resistivity x (L/A), where A is the area of the uniform cross section, and L is the distance between the electrodes.
So, to bring the resistance down to something within the range of you multimeter's accuracy, simply construct the measuring set up to have A >> L, then do the math from your resistance measurement to get back to resistivity.  For instance, if you pick a circular cross seciton, with D = 2r = 20 cm (about 8 inches), then A = 315 cm^2.  If you have 1.5 cm of liquid between the two plates, then A/L = 210.  If your meter reads 100K Ohms, then you know the resistivity is 210 x 100K = 21M Ohms.
By the way, the conductivity I found for "ultra-pure" water is 18.2M (Ohm x cm).  I forget if that is before or after exposure to air.  The CO2 in the air dissolves into the H2O, changing its Ph and conductivity.  Don't be surprised, if you are working with highly purified water, when its conductivity changes after exposure to air.
A: If you're looking to measure conductivity on a budget, try the project that this person created.
It includes a good explanation of how to do it, plus the circuit and some other goodies.
A: I entirely agree with @Vintage in his offered answer to @BarsMonster's question. It seems to me that the question isn't how to make the most accurate, error free measurement for water. Rather, what's a practical method that can give a 1st order, believable result? Placing two clean metal plates reasonably close together where d/A >> 1 and then measuring the resistance using a quality DMM is a practical solution. If d/A >> 1 then second order fringe effects can be reduced to reasonable levels and a 'fairly accurate' and reproducible result can be obtained. 
In the special case of this measurement, in already relatively clean water, as long as the two metal plates are reasonably clean and oil/grease free the measurement should be fairly consistent. Remember, you're about to make a measurement up in the MOhm/cm range (18.2MOhm/cm maximum). A a few hundred ohm/cm this way or that doesn't really make a significant difference. 
Furthermore, if you happen to have a sample of water with known conductivity, you can generate a calibration curve for your home built conductance meter. For example, most chemistry departments have water purifiers with conductance meters on them. Production of 18.2 MOhm/cm water is routine. No doubt they'd share a few hundred mL. If necessary, you can even formulate a series of salt solutions of known concentration (and therefore also known conductivity) and further extend the calibration curve. 
Of course, such an approach won't be perfect or ideal. But it will be practical and very cheap. As long as you keep its limitations in mind, I would say it's a reasonable approach. 
A: That DI water has some specific resistance, not resistance! 
To measure that, You need a apropriate probe, inserting 
the tips of a vanilla ohmmeter into the water, excuse, 
is ridiculous! Conductivity measurements have to be made using 
alternating currents. The frequency is about some kHz to some dozen kHz. 
For this purpose You can buy special meters, called conductometers, 
(In electrochemistry one deals with conductance, not resistance) 
at a range of cheap to luxurious. 
A: Two things come to mind. One a matter of experimental technique, and the other one of definitions.


*

*Firstly: how clean were/are the probes? If you introduce a water soluble material from the surface of the probes you will have true screwed up your measurement.

*Secondly: Are you sure that value is not for the resistivity of water. Resistivity is 
generally symbolized with $\rho$ and is related to the resistance by
$$ R = \int \frac{\rho}{A} dr $$
where $A$ is the cross-sectional area presented along the path element $dr$. This 
implies that the resistance is dependent on the geometry of the measurement. Just 
plunging the probes into a glass of water isn't going to cut it.
The units you give are proper for resistivity, and not for resistance.
A: Try instead measuring the conductivity. You'll need to use a conductivity meter. Once you know the conductivity, you'll know the resistivity, as it is the reciprocal: $\rho = \frac{1}{\sigma}$
You might want to take a look at this thread.
