Abstract. Measurements of ULF electromagnetic disturbances were carried out in Japan before and during a seismic active period (1 February 2000 to 26 July 2000). A network consists of two groups of magnetic stations spaced apart at a distance of ≈140 km. Every group consists of three, 3-component high sensitive magnetic stations arranged in a triangle and spaced apart at a distance of 4–7 km. The results of the ULF magnetic field variation analysis in a frequency range of F = 0.002–0.5 Hz in connection with nearby earth-quakes are presented. Traditional Z/G ratios (Z is the vertical component, G is the total horizontal component), magnetic gradient vectors and phase velocities of ULF waves propagating along the Earth’s surface were constructed in several frequency bands. It was shown that variations of the R(F) = Z/G parameter have a different character in three frequency ranges: F₁ = 0.1 ± 0.005, F₂ = 0.01 ± 0.005 and F₃ = 0.005 ± 0.003 Hz. Ratio R(F₃)/R(F₁) sharply increases 1–3 days before strong seismic shocks. Defined in a frequency range of F₂ = 0.01 ± 0.005 Hz during nighttime intervals (00:00–06:00 LT), the amplitudes of Z and G component variations and the Z/G ratio started to increase ≈ 1.5 months before the period of the seismic activity. The ULF emissions of higher frequency ranges sharply increased just after the seismic activity start. The magnetic gradient vectors (∇ B ≈ 1 – 5 pT/km), determined using horizontal component data (G ≈ 0.03 – 0.06 nT) of the magnetic stations of every group in the frequency range F = 0.05 ± 0.005 Hz, started to point to the future center of the seismic activity just before the seismoactive period; furthermore they continued following space displacements of the seismic activity center. The phase velocity vectors (V ≈ 20 km/s for F = 0.0067 Hz), determined using horizontal component data, were directed from the seismic activity center. Gradient vectors of the vertical component pointed to the closest seashore (known as the "sea shore" effect). The location of the seismic activity centers by two gradient vectors, constructed at every group of magnetic stations, gives an ≈ 10 km error in this experiment.